UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Analysis of DNA uptake biases in bacteria with and without uptake specificity Mora Rodríguez, Marcelo Andrés 2020

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2021_may_mora_marcelo.pdf [ 14.67MB ]
Metadata
JSON: 24-1.0395493.json
JSON-LD: 24-1.0395493-ld.json
RDF/XML (Pretty): 24-1.0395493-rdf.xml
RDF/JSON: 24-1.0395493-rdf.json
Turtle: 24-1.0395493-turtle.txt
N-Triples: 24-1.0395493-rdf-ntriples.txt
Original Record: 24-1.0395493-source.json
Full Text
24-1.0395493-fulltext.txt
Citation
24-1.0395493.ris

Full Text

  ANALYSIS OF DNA UPTAKE BIASES IN BACTERIA WITH AND WITHOUT UPTAKE SPECIFICITY  by   Marcelo Andrés Mora Rodríguez   M.Sc. University of Northern British Columbia, 2012 Licentiate, Pontificia Universidad Católica del Ecuador, 2006      A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY   in    THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Zoology)     THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)    December 2020 © Marcelo Andrés Mora Rodríguez, 2020   ii The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled: Analysis of DNA uptake biases in species with and without uptake specificity  submitted by Marcelo Mora  in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Zoology  Examining Committee: Rosemary Redfield, Zoology Supervisor  Matthew Pennell, Zoology Supervisory Committee Member  Sarah Otto, Zoology University Examiner Steve Hallam, Microbiology and Immunology University Examiner     iii Abstract DNA uptake is the first step in natural transformation of bacteria, leading to DNA internalization and recombination. It is, therefore, a key determinant in genome evolution. Most bacteria take up DNA indiscriminately, but in Pasteurellaceae and Neisseriaceae, the uptake machinery binds preferentially to short sequences enriched in their genome, called DNA uptake signal sequences (USS or DUS). This enrichment is responsible for bacteria preferentially taking up DNA from close relatives, called ‘self-specificity’. My study’s goal was to characterize factors influencing uptake bias across the genome (chapter 2), as well as determined if uptake biases were present in a species without self-specificity (chapter 3).   Chapter 2 describes my genome-wide analysis of DNA uptake by Haemophilus influenzae, a species with strong uptake bias, using both measured uptake and the predictions from a simulation model of DNA uptake. Genomic maps of DNA uptake were developed by recovering DNA after it was taken up and deep sequencing it, using DNA preparations sheared into small (50-800 bp) and large (1.5-17 kb) fragment sizes. For short donor DNA fragments, I found a strong uptake bias, up to 1000-fold, that was proportional to the similarity of the USSs to the consensus, including in DNA topology features. I found no evidence of an effect on uptake of non-USS sequences. Uptake of large donor DNA fragments had much less variation than short fragments, with 90% of genomic positions having uptake within 2-fold of the mean. However, it was difficult to assess the USS-dependent factors responsible for this variation because of stochastic noise arising from intrinsic biases of the DNA sequencing process.   Chapter 3 describes an analysis to determine if uptake biases are present in Acinetobacter baylyi, a species without self-specificity. This was done by sequencing a degenerate region of synthetic DNA fragments, recovered after uptake by competent A. baylyi cells. Recovered fragments had the same sequence distribution as input DNA, which suggests that there is no evidence of uptake biases.     iv Lay Summary  Some bacteria can take up DNA from the environment. A few species of these bacteria prefer DNA fragments that have a short sequence pattern, called uptake sequences. These sequences are found at higher frequency than expected by chance in their own genomes and as a consequence, these bacteria preferentially take up DNA from their own species. The first goal of this study was to identify the factors responsible for the differences in uptake of regions in the genome of Haemophilus influenzae, a species that prefers its own DNA. I found that uptake depended strongly on the sequence, distribution, and DNA shape of uptake sequence. The second goal was to determine if uptake preferences are found in all species or only the ones that prefer their own DNA. No preferences were detected in uptake of DNA fragments taken up by Acinetobacter baylyi, a species that can take up DNA from different species.                v Preface Chapter 1, 3 and 4 were written by Marcelo Mora (MM). Chapter 2 was written by MM and Rosie Redfield (RR) with suggestions from Joshua Mell Chang (JM), Garth Ehrlich (GE), and Rachel Ehrlich (RE). Chapter 2: A version of this material has been published in the iScience journal as:  Mora, M.; Mell Chang, J.; Ehrlich, G. D.; Ehrlich, R.L.; Redfield, R.J.  Genome-wide analysis of DNA uptake by naturally competent Haemophilus influenzae. ISCIENCE (2021), doi: https://doi.org/10.1016/j.isci.2020.102007  Experimental design was conceived by JM and RR. MM conducted the uptake experiments. Sequencing of recovered and input DNA was done by JM. Sequencing of the Haemophilus influenza PittGG genome was done by GE and RE. Alignment of NP recovered DNA was done by JM. Alignment of PittGG was done by MM with scripts from JM.  USS scores were calculated by MM using scripts from JM. Predictive uptake models were designed and scripted by MM and RR. Figure 2.3, 2.4, 2.15A were made by RR. The rest of figures and bioinformatic analysis were made by MM, with guidance from RR. Chapter 3: Experimental design was conceived by JM and RR. MM conducted the uptake experiments. Synthetic fragments were designed by MM, with guidance from RR and JM. Sequencing was done by the UBC Bioinformatic consortium. Bioinformatic analysis were designed and done by MM, with guidance from RR.         vi Table of Contents Abstract ................................................................................................................................. iii Lay Summary ......................................................................................................................... iv Preface .................................................................................................................................... v Table of Contents ................................................................................................................... vi List of Tables .......................................................................................................................... ix List of Figures .......................................................................................................................... x List of Abbreviations ............................................................................................................. xiv Acknowledgements ............................................................................................................... xv Dedication ............................................................................................................................ xvi CHAPTER ONE ......................................................................................................................... I General Introduction ............................................................................................................... I What is natural transformation and why is it important? ................................................................. I DNA uptake as part of the natural transformation process .............................................................. 2 Self-specificity and uptake sequences .............................................................................................. 4 Protein-DNA interactions ................................................................................................................. 9 Main questions about uptake bias and self-specificity .................................................................... 10 CHAPTER TWO ...................................................................................................................... 13 Genome-wide analysis of DNA uptake across the outer membrane of naturally competent Haemophilus influenzae ........................................................................................................ 13 Introduction ................................................................................................................................... 13 Direct measures of DNA uptake bias: ................................................................................................................ 15 Evolution of uptake sequences in the genome: ................................................................................................ 16 Methods ......................................................................................................................................... 18  vii Results ........................................................................................................................................... 24 A computational model of DNA uptake ............................................................................................................. 24 Model results: .................................................................................................................................................... 27 Generation of experimental DNA uptake data .................................................................................................. 30 Removal of contaminating Rd DNA: .................................................................................................................. 31 Uptake ratios: .................................................................................................................................................... 32 Sensitivity is limited by low sequencing coverage: ............................................................................................ 38 Uptake ratios show no periodicity across the genome: .................................................................................... 40 Uptake of short-fragment 86-028NP DNA ......................................................................................................... 42 DNA shape effects: ............................................................................................................................................ 47 Uptake of fragments with more than one USSs ................................................................................................ 50 Uptake of long-fragment 86-028NP DNA .......................................................................................................... 51 Prediction of uptake of PittGG DNA .................................................................................................................. 54 Predicted competition with other DNAs in the human respiratory tract .......................................................... 57 Discussion ...................................................................................................................................... 60 Implications for DNA uptake .............................................................................................................................. 60 Implications for genetic exchange ..................................................................................................................... 61 Competition with DNAs from human cells and other respiratory bacteria: ...................................................... 63 LIMITATIONS OF THE STUDY .............................................................................................................................. 63 CHAPTER THREE .................................................................................................................... 65 Analysis of DNA uptake bias in Acinetobacter baylyi, a bacterium without uptake specificity 65 Introduction ................................................................................................................................... 65 Materials and methods ................................................................................................................... 67 Strains and growth conditions ........................................................................................................................... 67 Natural transformation ...................................................................................................................................... 68 Synthesis of double-stranded input DNA fragment ........................................................................................... 68 DNA uptake assays ............................................................................................................................................. 69  viii Sequencing and library preparation .................................................................................................................. 70 Bioinformatics .................................................................................................................................................... 70 Motif finder analysis .......................................................................................................................................... 72 Results and Discussion .................................................................................................................... 72 Recovery of DNA ................................................................................................................................................ 72 PCR artefacts and quality controls ..................................................................................................................... 73 Sequence enrichment ........................................................................................................................................ 75 CHAPTER FOUR ..................................................................................................................... 84 Conclusions ........................................................................................................................... 84 Evolution of uptake bias and molecular drive ................................................................................. 85 Next steps, what remains to be understood ................................................................................... 88 Bibliography .......................................................................................................................... 91 Appendix ............................................................................................................................. 104  ix List of Tables Table 2.1. Bacterial strains used in this study _______________________________________________________ 20 Table 2.2.  Frequencies of Hin-type USS in genomes of other species ____________________________________ 57 Table 2.3. Predicted relative uptake1 of H. influenzae DNA in simulated competition with DNAs of other species2. 59 Table 3.1. Number of sequenced reads for each sample, as well as the percentage of reads with unique degenerate regions _____________________________________________________________________________________ 72 Table 3.2. Probability in percentages of having at least one of a 3-8 bp long sequence in the 69 bp degenerate region and 131 bp fixed region of the fragment. Percentages were calculated, based on the GC content of synthetic fragments, for kmers with A or Ts as well as C or Gs. _________________________________________________ 81 Table S1.  Sample metadata ___________________________________________________________________ 104 Table S2 uptake-prediction position-specific scoring matrix from Mell et al.’s degenerate-sequence uptake experiment _________________________________________________________________________________ 106              x List of Figures Figure 2.1. The H. influenzae uptake signal sequence. A. Sequence logo showing the individual contributions to genomic abundance of bases in the USS motif (Maughan 2010). B. Sequence logo showing the individual contributions to uptake of bases in the USS motif, as measured by Mell et al. (2012). C. Conserved USS segments. _ 16 Figure 2.2. Frequency distribution of USS scores for all positions in H. influenzae and random-sequence genomes. Related to Figure 1. 86-028 NP (blue), PittGG (orange), Rd (gold), and four random-sequence 1.9 Mb genomes with the same base composition (38%G+C, black and grey).  Scores were calculated with the uptake scoring matrix in Table S2. The numbers in the lower right are the numbers of positions meeting cutoff scores of 9.5, 10.0 and 10.5 bits. Inset: Expanded view for positions with scores higher than 9 bits. __________________________________ 26 Figure 2.3. A computational model to predict DNA uptake. A. Components of the DNA uptake model (see Methods and Results for details). B. & C. Model predictions for uptake centered at a 12-bit USS for: B. 100, 200, and 300 bp fragments, C. a mixed distribution of fragments between 25-300 bp with and without baseline uptake. D. E. & F. Model predictions for uptake of a 3000 bp region with 3 USSs (red squares, scores in black) using different fragment-length distributions: D. 50-300 bp fragments, E. 50-2000 bp fragment, F. 1-14 kb fragments. G. & H. Predicted DNA uptake of a 50 kb segment of the 86-028NP genome using G. short and H. long fragment-length distributions: I. Locations and scores of USS10s in this 50 kb segment. ___________________________________ 29 Figure 2.4: Distributions of fragment lengths in input DNA preparations. Relative abundances of DNA fragment lengths were estimated from Bioanalyzer data for input DNA samples from strains 86-028NP (blue) and PittGG (orange).  A. Short-fragment preparations. B. Long-fragment preparations. Solid lines: length distributions of fragments in input DNA preparations, normalized to most frequent length.  Dashed lines: Length distributions of sequenced fragments, with arbitrary scaling. Insets: Bioanalyzer pseudo-gel images of sheared DNAs (NP: 86-028NP, GG: PittGG). Bioanalyzer molecular weight markers are shown in purple and green. _________________ 30 Figure 2.5 Locations of positions with missing uptake ratio data. Each point represents a genome position for which an uptake ratio could not be calculated. The points are vertically jittered, so segments with no coverage appear as black rectangles.  A. 86-028NP, short-fragment data.  B. 86-028NP, long-fragment data. C. PittGG, short-fragment data. D. PittGG, long-fragment data. _____________________________________________________________ 32 Figure 2.6. Experimentally determined uptake ratios for a 50 kb segment. The X-axis is the same 50 kb segment of the 86-028NP genome as Figure 2.3G and H. Grey points indicate positions with input coverage lower than 20 reads. Gaps indicate unmappable segments. A. Uptake ratios of short-fragment DNA. Inset: Same data with a logarithmic-scale Y-axis. B. Uptake ratios of long-fragment DNA. C. Locations and scores of USS10s. ______________________ 34 Figure 2.7. Experimentally determined uptake ratio maps and USS10 maps. Grey points indicate positions with input coverage lower than 20 reads.  Gaps indicate unmappable positions.  A-C:  Maps of a 50 kb segment of the PittGG genome.  A. Uptake ratios of short-fragment PittGG DNA.  B. Uptake ratios of long-fragment PittGG DNA.  C. PittGG USS10 positions and scores.  D-I: Whole-genome maps.  D. Uptake ratios of short-fragment 86-028NP DNA.  E.  xi Uptake ratios of long-fragment 86-028NP DNA.  F.  86-028NP USS10 positions and scores.  G. Uptake ratios of short-fragment PittGG DNA.  H. Uptake ratios of long-fragment PittGG DNA.  I.  PittGG USS10 positions and scores. ___ 36 Figure 2.8.  Shapes of typical 86-028NP uptake peaks. Blue dots: uptake ratios after smoothing with a 31 bp window. A.-H.: Orange dots: uptake ratios without smoothing (note that Y axis is offset by 0.5 units). Red triangles and numbers: locations and scores of USS.  A.-F. 86-028NP short-fragment DNA: A. and B.: peaks at strong USS. C. and D.: peaks at weak USS. E. and F.: peaks at pairs of USS separated by A. 69 bp and B. 230 bp. G. and H. Uptake ratio spikes not at USS in 86-028NP long-fragment DNA. I. and J.  Purple dots: sequencing coverage of input 86-028NP long-fragment DNA. _____________________________________________________________________ 37 Figure 2.9.  Variation in read coverage. Read coverage of the 86-028NP long-fragment (green) and short-fragment (purple) input samples over a 50 kb genome segment. _______________________________________________ 38 Figure 2.10.  Shape analysis of isolated USS10 peaks. Short-fragment uptake ratio data for positions around 158 86-028NP USS10s that were separated by at least 1200 bp from other USS10s and had uptake ratios of at least 3.0. ___________________________________________________________________________________________ 39 Figure 2.11. Depths of uptake ratio valleys with and without contamination correction. Uptake ratios of positions in the 86-028NP short-fragment dataset that were at least 1kb from the nearest USS9.5 and whose uptake coverages were either corrected for contamination with homologous Rd sequences (left, 272,547 positions) or not corrected for contamination because they had no Rd homology (right, 89,418 positions). Orange and red dashed lines indicate valley uptake predicted by the original and revised models respectively. _________________________________ 40 Figure 2.12. Tests of periodicity. Fourier-transform analyses were performed using R-package RCA.  The X-axes are log10 of repeat period in bp; Y-axes are log10 of the relative periodicity at each repeat period.  A-D.  Tests using coverage in input samples.  E-H. Tests using uptake ratios. Samples: A & E: 86-028NP short fragments; B & F: 86-028NP long-fragments; C & G:Pitt GG short fragments; D & H: PittGG long fragments. _____________________ 41 Figure 2.13. Predicted and observed DNA uptake analysis for short fragments of 86-028NP DNA. A. Map of uptake ratios and initial model predictions. The blue points show the same uptake ratio map as in Figure 2.6A. The orange points show the same predicted uptake as in Figure 2.3G. B. Map of differences between observed minus expected uptake ratios. C. Locations and scores of USS9.5s. ___________________________________________________ 43 Figure 2.14. Short-fragment uptake ratios and predicted DNA uptake at isolated 86-028NP USS as a function of USS score. Predicted or measured DNA uptake at 209 USS9.5 positions separated by at least 1000 bp from the nearest USS10s. The blue dots show the measured uptake ratios; grey bars show the ranges of the three replicates at each position. The blue line shows a sigmoidal function fit to these points. The small orange dots and line show uptake predicted by the original model at the same positions, and the small purple and red dots show uptake predicted by the intermediate and revised model versions discussed in the text. _____________________________________ 44 Figure 2.15. Effects of within-USS interactions on USS scores. USS scores calculated with and without interactions effects for 209 isolated 86-028Np USS9.5s (see Methods for details). Red line shows expected scores if the  xii interactions had no effect. Point color indicates the uptake ratios at the USS: yellow, <0.01; red, 0.01-0.1; purple, 0.1-0.5; blue 0.5-1.5; green, >1.5. ________________________________________________________________ 47 Figure 2.16. Predicted shape features of USS. The thick grey line reproduced in each box shows shape analysis of the consensus USS sequence. Blue and orange lines in each box show shape analysis of genomic USS separated by at least 500 bp, grouped by score and coloured by uptake ratio. The orange and yellow bars below each box indicate components of the USS (see Figure 1): light orange: outer core; dark orange: inner core; yellow: AT tracts. Panels A-D USS9.5-10: Blue: uptake ratios <0.2 (n=68, mean USS score =9.7). Orange: uptake ratios >0.2 (n = 10, mean USS score =9.8). Panels E-H: USS10.0-10.5: Blue: uptake ratios <0.6 (n=47, mean USS score=10.22). Orange: uptake ratios >2.0 (n=10, mean USS score=10.26). Panels I-L: USS10.5-11.0: Blue: uptake ratios <0.6 (n=14, mean USS score=10.64). Orange: uptake ratios >2.0 (n=59, mean USS score=10.79). A, E and I. Minor groove width, in Å. B, F and J. Propeller twist, in degrees. C, G and K. Helix twist, in degrees. D, G and L. Base pair roll, in degrees. See also Figure S5. Grey bars for each point show the standard error.  Dots indicate significant differences by Kolmogorov-Smirnov between high-uptake and low-uptake positions with (red dots) and without (black dots) Bonferroni correction. __________________________________________________________________________________ 49 Figure 2.17.  Analysis of DNA uptake effects of USS10 pairs in the 86-028NP genome. Frequencies of spacings between close USS10 pairs. B. Uptake ratios for isolated USS10 (blue points, data from Figure 2.14) and for centers of pairs of USS10 whose centers are 14-100 bp apart (red points) or 0-13 bp apart (green points). _______________ 51 Figure 2.18. Predicted and observed uptake of long 86-028NP DNA fragments. A. and C. Uptake maps for 86-028NP long-fragment DNA. Orange points: USS-dependent uptake predicted by the revised model. Blue points: mean uptake ratios from 3 replicate experiments (grey indicates input coverage <20 reads, gaps indicate unmappable positions). B. and D. USS9.5 positions and scores. A. and B. The same 50 kb genome segment shown in previous figures. C. and D. Whole genome. ________________________________________________________________ 52 Figure 2.19. Simulated noise analysis. A. Effects of added red noise on simulated noise-free coverage. Black points: no added noise; Green, yellow and red points: simulated noise added with multipliers of 1.0, 2.0 and 3.0 respectively. B. Correlation coefficient from simulated uptake ratios with and without different levels of noise. X-axis represents the multiplicative factor applied to the coverage-dependent amount of noise added. Simulated results for 86-028NP-short and 86-028NP-long DNA fragments are shown in blue and red, respectively. Dashed lines at 0.93 (blue) and 0.60 (red) indicate the real-data correlation of predicted uptake with observed uptake ratios for 86-028NP-short (top) and long (bottom) fragment size distributions, respectively. _________________________ 53 Figure 2.20 Predicted and observed uptake ratios for PittGG short and long DNA fragments.  A and E: Uptake maps for the first 50kb of the PittGG genome; C. and G.: Uptake maps for the full PittGG genome. B., D., F. and I.: Vertical tick marks indicate locations and scores of USS10s.  Orange lines: USS-dependent uptake predicted by the revised model.  Blue lines: Mean uptake ratios from 3 replicate experiments (grey points indicate positions with <20 reads input coverage, gaps indicate unmappable positions). Note that some grey dots are beyond the tops of some panels. _____________________________________________________________________________________ 57  xiii Figure 3.1: Diagram of synthetic fragments used for the uptake experiments. First, two oligos were annealed (A) and extended (B). Each region of the fragment is color coded and the sequence of each oligo is shown in (C). Red and purple are the Illumina sequencing primer binding region. Black are the 3 fixed bases that were used to measure synthesis errors (see Methods). Green are the 31 bp fixed bases region. Orange shows the annealing area between the two oligos.  In grey, the 69 bp degenerate bases region is shown. ____________________________ 69 Figure 3.2. Incorrect base calling analysis. A. the frequency incorrect base calling at a 61 bp section of the fixed region of the fragments of the input (red) , 3 replicates (green, blue and purple) and the no-uptake control (olive) dataset. B. the frequency of undefined bases, called as 'N', among all datasets. ___________________________ 74 Figure 3.2 Comparison between the frequency of artificially enriched 5mers in the input dataset, with low, medium and high frequency, with their original kmer frequencies. A. distribution of the input 5mers. Red arrows showed the frequency of the 5mers chosen to be artificially enriched. B-D. P-value of chi-square analysis of 5mers with low (B), medium (C) and high (D) frequency with and without having their frequency enriched by 0.02 to 0.3-fold (x-axis). 76 Figure 3.3: Normalized frequency of 5mers in the 69 bp degenerate region from the recovered (y-axis) divided by the input (x-axis) fragments. 5mers from the Acinetobacter BD413 recovered replicates are shown in A-C, while 5mers from Acinetobacter no-uptake control (T205) is shown in D. _____________________________________ 78 Figure 3.4 Control sequence-motif used to determine sensitivity of MEME and DREME algorithms. A. sequence motif used to replace random 8mers of the input dataset to create a positive control dataset. B. Table showing the sensitivity of MEME and DREME algorithms when these algorithms were used to detect the Motif in A. when this was replaced in 10.11 to 26.95% of the input reads. E-value, as described by MEME-suite, is equal to the P-value calculated using Fisher exact test times the number of candidate motifs tested. ___________________________ 79 Figure 3.5: Analysis of the 3 motifs with highest E-values in recovered fragments replicates and the no-uptake control with the DREME algorithms. ______________________________________________________________ 80            xiv List of Abbreviations DUS = DNA uptake sequence (terminology traditionally used in literature for uptake sequences Neisseriaceae) GG = Haemophilus influenzae PittGG strain NP = Haemophilus influenzae 86-028NP strain USS = Uptake Signal Sequence (terminology traditionally used in literature for uptake sequences Pasteurellaceae)             xv Acknowledgements I would like to give a special thanks to my supervisor Dr. Rosemary Redfield for all her guidance, financial support, and advice. I would also like to thank my supervisory committee: Dr. Matthew Pennell, Dr. Laura Parfrey, and Dr. Tom Beatty for all their suggestions and feedback. To Dr. Erin Gaynor for her feedback during the initial steps of the project. To Dr. Rachel Simister for her assistance with the bioanalyzer fragment analysis and for the statistical advice. To Lauri Lintott and Scott Mastromatteo for their assistance, advice, and friendship. To Dr. Beate Averhoff and Dr. Klaus Harms for their assistance in providing Acinetobacter baylyi strains. To Dr. Rachel Enrich and Garth Enrich for their collaboration and suggestions regarding the second chapter. Finally, I would like to thank Dr. Joshua Mell Chang, whose advice and feedback was essential for this project's success.                  xvi Dedication To my wife whose sacrifice, and support made this possible. To my parents and my sister who believed in me from the start until the end. To my boys Gabriel and Leonardo, you are the light in my path.                     I CHAPTER ONE General Introduction What is natural transformation and why is it important?  Bacteria have three main horizontal gene transfer mechanisms that incorporate foreign genetic material into their genomes, conjugation, transduction and natural transformation (Thomas and Nielsen, 2005). Unlike conjugation and transduction, transformation is fully controlled by the recipient cell. During transformation  competent cells take up DNA and recombine it into their own genome. Like the other mechanisms, natural transformation is a mechanism that generates variation that can promote adaptive evolution. Some adaptive traits that have been reported to be acquired by natural transformation and that promote survival and spread of pathogenic bacteria include antibiotic resistance genes, virulence genes, and capsular switching events (Domingues et al., 2012; Straume et al., 2015).  Recombination is an integral part of the natural transformation process, and a large number of recombination events have been found in bacterial genomes (Gonzalez-Torres et al., 2019; Smith et al., 1991). Recombination of DNA fragments during natural transformation can generate diversity and change the fitness of bacteria. For instance, changes in antigenic type of pilin proteins that help Neisseria gonorrhoeae binding to the host epithelia can arise by recombination, after natural transformation by DNA containing pilE gene from dead cells and recipient cells with the pilS gene (5). This is why some researchers have compared adaptive variation generated by natural transformation to the variation generated by meiotic sex. Both types of recombination can decrease the accumulation of slightly deleterious mutations (Kondrashov, 1982; Pamilo et al., 1987), which is one of the main disadvantages of asexual reproduction (Muller, 1932). Additionally, both can generate genetic variation that will allow higher fitness landscapes to be explored by natural selection (Vos, 2017).  The evolutionary function of natural transformation remains controversial. For the reasons mentioned above, it has been proposed that natural transformation evolved to generate variation (Smith et al., 1991). One weakness of this hypothesis is that the role of the highly2  conserved recombination machinery is to be part of the bacterial recombinational repair machinery and not to generate variation by recombination (Bernstein et al., 1985; Cox et al., 2000; Rocha et al., 2005; Smith, 1985) . Another weakness of this hypothesis is that because competent cells take up DNA from dead cells, the DNA could be carrying deleterious mutations from dead bacteria (Redfield, 1988). Although deleterious mutations are considered to be purged from a population by purifying selection, in a naturally competent populations these mutations remain in DNA to be recombined into other cells. Another cost of recombination, one shared with meiotic sex, is the breakup of adaptive allele combinations by recombination, known as the recombination load (Otto and Lenormand, 2002; Vos, 2017).  As an alternative to the generating-variation hypothesis, researchers have proposed that natural transformation has evolved for DNA repair or for the acquisition of nucleotides for nutrients (Michod et al., 1988; Redfield, 1993a). The transformation-for-repair hypothesis builds on the role of the recombination machinery in recombinational repair and is supported by Michod et al. (Michod et al., 1988) experiments finding that natural transformation increases with UV dosage in Bacillus subtilis. However, natural competence is tightly regulated, and Redfield (Redfield, 1993b) showed that it is not induced by DNA damage, suggesting that DNA-repair is not its selected function. In contrast, nutritional stress did induce competence in Haemophilus influenzae, B. subtilis, and Vibrio cholerae (Jarmer et al., 2002; Meibom et al., 2005; Redfield, 1993a; Weinrauch et al., 1990), which supports the hypothesis that transformation evolved for nutrient acquisition. Additionally, Acinetobacter baylyi strains with DNA uptake showed higher fitness than strains without DNA uptake (Hülter et al., 2017). This fitness advantage was independent of recombination and of sequence similarity between donor and recipient, which is inconsistent with both genetic variation and the DNA repair hypothesis (Hülter et al., 2017). DNA uptake as part of the natural transformation process  Transformation in gram negative bacteria has three steps, uptake across the outer membrane, translocation across the inner membrane, and recombination. During the uptake step, competent species use a type IV pilus mechanism to take up extracellular DNA through a secretin pore in the outer membrane  (Schwarzenlander and Averhoff, 2006; Stone and Kwaik, 1999). In 3 gram positive bacteria a similar mechanism is used to bring DNA across the cell wall. In the translocation step, one DNA strand is moved from the periplasm to the cytoplasm while the other strand is degraded (Krüger and Stingl, 2011). In the recombination step, internalized DNA fragments which are highly similar to the recipient chromosome, undergo RecA-mediated recombination; while others, that do not share high similarity to the recipient chromosome, are degraded (Maughan et al., 2008).    Figure 1.1: Uptake and translocation machinery of Haemophilus influenzae. A. Type 4 pili proteins (in green) pull a dsDNA fragments through the outer membrane into the periplasm. Pre-pilin and pilin subunits are in dark green. Proteins involved in translocation of one strand of DNA to the cytoplasm and degradation of the other strand are in yellow.  Numbers indicate functional categories of the uptake machinery: 1. Pseudopili/pili, 2. Prepilin processing leader peptidases, 3. Traffic NTPases, 4. Pilus assembly proteins, and 5. Secretins. Donor DNA is represented in black, while recipient genome is represented in red. ComEComE1PilDPilBRec2ComFNupC GPilCNucleotides12534Outer membraneInner membraneCytoplasmChromosome4 DNA uptake machinery is highly conserved and comprises proteins that can be functionally subdivided into 5 classes: pilins, prepilin processing leader peptidases, traffic NTPases, pilus assembly proteins, and an outer membrane channel pore (Figure 1.1) (Chen and Dubnau, 2004; Korotkov et al., 2011). Type 4 pilin subunits are small proteins that interact with each other forming the pilus fiber. These proteins arise from precursors that have an N-terminal leader peptide cleaved by a prepilin processing leader peptidase allowing export and maturation of the pilin subunits (Giltner et al., 2012; Korotkov et al., 2011). Once exported to the periplasm, pilin is assembled by an interaction of pilus assembly proteins and traffic NTPases that use ATP to polymerize (and later depolymerize) the pilin subunits. DNA binds to the pilus filaments that extend through the pore in the outer membrane. After binding, DNA is pulled through the pore by the retraction force generated when the NTPases depolymerizes the subunits (Ellison et al., 2018). DNA internalization is aided by a periplasm ratchet protein (Ellison et al., 2018; Hepp and Maier, 2016). Inside the gram-negative periplasm, one DNA strand is translocated to the cytoplasm through an inner membrane channel protein (Rec2) (Draskovic and Dubnau, 2005), while the other strand is degraded (Krüger and Stingl, 2011). How DNA is pulled thought the secretin pore into the bacterial periplasm is not well understood since very few DNA binding receptors have been described in competent species. In Neisseria, the DNA binding receptor protein is a minor pilin (Berry et al., 2016). No homologs have been described in species outside Neisseriaceae, including in H. influenzae where no DNA binding receptor has been identified yet, even though it is presumed to be also a minor pilin.  In V. cholerae the DNA binding receptor has been located at the tip of the pilus filament (Ellison et al., 2018) although the gene has not been identified. However, in Streptococcus pneumoniae DNA can bind anywhere along the pilus length (Laurenceau et al., 2013). These results suggest that the DNA binding protein is not well conserved among transformable species. Self-specificity and uptake sequences Natural transformation is present in most bacterial phyla (Figure 1.2) with some, such as Gama-proteobacteria and Firmicutes including several transformable representatives. Other phyla that 5 have transformable representatives include Deinococcus, Chlorobi, Cyanobacteria, Alpha, Beta, Epsilon, and Gamma-proteobacteria, Actinobacteria (Figure 1.2).  Most naturally competent bacteria take up DNA from any source. However, members of Pasteurellaceae, Neisseriaceae, and Campylobacter jejuni have strong preferences for taking up DNA from their own or closely related species (red dots in Figure 1.2). (Dougherty et al., 1979; Levine et al., 2007; Palmen et al., 1993; Sisco and Smith, 1979; Wang and Taylor, 1990). Early studies in Pasteurellaceae and Neisseriaceae have shown that these species have strong preferences for taking up DNA fragments with short species-specific sequences. These sequences are highly overrepresented in the genome fragments that this species prefer to take up, explaining the self-specificity. These sequences are called uptake signal sequences or USS in Pasteurellaceae and DNA uptake sequences or DUS in Neisseriaceae.   6  Figure 1.2 Maximum-likelihood phylogenetic tree from 31 concatenated conserved protein genes from DY Wu et al. (Wu et al., 2009). Black dots indicate clades containing at least one species with reported natural transformation. Red dots indicate clades with at least one species with uptake sequences and self-specificity. Orange stars show the locations of the clades to which the species used in the present study (H. influenzae and A. baylyi) belong.    7 Uptake sequences, an 11-bp USS (AAGTGCGGTCA) and a 10-bp DUS (GCCGTCTGAA), were initially identified by sequencing cloned DNA fragments that have been preferentially taken up by H. influenzae (an 11-bp USS (AAGTGCGGTCA)) and by Neisseria gonorrhoeae (a 10-bp DUS (GCCGTCTGAA)) (Danner et al., 1980)(Goodman and Scocca, 1988). Later studies showed that the USS also includes two AT-rich regions (Fitzmaurice et al., 1984; Smith et al., 1995, 1999) (Figure 1.3). The DUS does not have AT-rich regions, but two extra 3’ bases that increase transformation efficiency approximately 3-fold (Ambur et al., 2007).    Figure 1.3  A. USS sequence logo based on 2206 genomic sequences in the H. influenzae genome (Maughan et al., 2010). Information content in bits has been scaled to account for the 38% G+C base composition of the H. influenzae genome. B.	Conserved USS segments. Sequencing of genomes from Haemophilus and Neisseria found that uptake sequences were distributed on average 1 every kb (Smith et al., 1999). It was also found that uptake sequences were not fixed but a variable family of sequences in which some bases were more conserved than others (Figure 1.3) (Redfield et al., 2006). Additionally, related species have a similar family of uptake sequences or dialects. For instance, within Pasteurellaceae two dialects of USS were found, one in species related to H. influenzae, ‘Hin-type USS’, and another in species related to Actinobacillus pleuropneumoniaea, ‘Apl-type-USS’ (Redfield et al., 2006). Similarly, in CoreInner coreAT-richsegmentAT-richsegmentB.8 Neisseriaceae, eight different uptake sequence motifs (12 nucleotides each) were identified, each differing from the rest by 2 to 4 bases (Frye et al., 2013). The self-specificity of DNA uptake by Pasteurellaceae and Neisseriaceae arises from the combination of strong uptake bias and the enrichment of preferred sequences in their genomes. The ultimate cause of uptake bias is not clear, but it might be related to the need for high binding affinity of the uptake receptor to the DNA (see below). However, the enrichment of the preferred sequences is well explained by an evolutionary process called molecular drive. Molecular drive is independent of natural selection; instead uptake sequences accumulate in genomes as a consequence of cycles of mutation, strong uptake bias and recombination (Danner et al., 1980; Maughan et al., 2010). Maughan et al. (Maughan et al., 2010) used a computer simulation model to show that this process can produce accumulation of uptake sequences with frequencies and distribution similar to those of real genomes. In this process, sequences change by random point mutations that alter their affinity for the uptake receptor (Maughan et al., 2010). Fragments with mutations with higher affinity for the receptor will then be preferentially taken up and recombined in the genome, and fragments with mutations that reduce affinity will be left behind. Uptake sequences will accumulate and improve over many rounds of mutation, biased uptake and recombination, reaching an equilibrium at which the gain of uptake sequences by recombination and loss of them by mutation is balanced (Maughan et al., 2010).  Molecular drive predicts that USSs will accumulate in genomic locations where they are not harmful to gene function. This is consistent with the low density of USS observed in coding regions, especially in highly conserved genes and in rRNA genes (Findlay and Redfield, 2009; Smith et al., 1999). Additionally, the gradual accumulation of uptake sequences by molecular drive is consistent with the low density of USS in DNA segments recently acquired from distantly related species, suggesting that uptake sequences accumulate after DNA fragments have been transferred (47).  Self-specificity in Pasteurellaceae and Neisseriaceae has been compared to eukaryotes mate-recognition in sexual species, which evolved to decrease the costs of recombining with distant relatives (Majewski, 2001; Treangen et al., 2008). However, similar selection for transformation 9 systems will be complex, and gradual accumulation of uptake sequences by molecular drive provides a much simpler explanation. Protein-DNA interactions DNA-binding proteins, in both Eukaryotes and Prokaryotes, will usually bind non-specifically at random sequences before finding their recognition sites. This is because their DNA-binding motif are relatively rare compared to the rest of the genome sequences.  Once bound to DNA by weak interactions, DNA-binding proteins scan for their preferred binding sequences using three main strategies: sliding or diffusion along the DNA, hopping between double helices, and intersegmental transfer (Halford and Marko, 2004; Redding and Greene, 2013).  Sliding proteins use non-specific DNA interactions to remain in contact with the DNA molecule while they scan for specific-binding sequences (Halford and Marko, 2004). In hopping, proteins attach and re-attach at another position. During intersegmental transfer, proteins move from one DNA position to a distant one by binding simultaneously to both when they are in contact (e.g. at a DNA loop). These mechanisms are not mutually exclusive, but intersegmental transfer is important only in proteins that bind as a dimer with two DNA-binding domains (Halford and Marko, 2004; Redding and Greene, 2013).  Both DNA sequence and DNA shape contribute to specific interactions that allow a protein to recognize its binding site. Sequence-specific interactions, called ‘sequence-readout’, occur when the amino acid side chains of protein recognition site contact the sides of the bases of the double stranded DNA (Rohs et al., 2010; Seeman et al., 1976). Proteins can also have electrostatic interactions between their positively-charged amino acids and negatively-charged phosphate groups of the DNA backbone (Cherstvy, 2009; Rohs et al., 2009), called ‘DNA shape-readout’. These interactions allow the protein to recognize the shape of the DNA backbone and may also improve interactions between amino acids and DNA bases (Gordân et al., 2013; Jones et al., 1999).  Uptake of DNA produces particular problems for DNA binding. One of these problems is the hydrodynamic resistance generated by pulling a long DNA molecule through an aqueous environment. Since H. influenzae does not use a free end to take up DNA, as evidenced by the 10 uptake of closed circular DNA molecules, DNA needs to be kinked before passing through the pore. If DNA binds to the tip of the pilus, both arms of the kinked molecule (~4 nm) will fit through the pore, but if DNA binds to the side of the pilus then the pore has to accommodate the DNA and the ~ 6 nm diameter pilus (Collins et al., 2005). Retraction of the type IV pilus by NTPases generates very strong forces (higher than 110 pN) that will require commensurately strong attachment of the DNA to the receptor (Clausen et al., 2009; Maier et al., 2002). Although the Neisseria DNA-binding receptor binds with high affinity to uptake sequences, the actual forces have not been measured (Berry et al., 2016). DNA uptake experiments using variant or degenerate uptake sequences showed that some positions make stronger contributions to uptake than others, suggesting that these are more important for DNA binding.  For instance, changes in the inner core GCGG motif of H. influenzae USS (Figure 1.2), and of the CTG motif of Neisseria meningitidis DUS, dramatically reduce uptake compared to changes at other positions (Berry et al., 2013; Frye et al., 2013; Mell et al., 2012). In H. influenzae, changing a core position close decreases uptake more than expected when this mutations occur in combination with mutations in the AT-rich regions (Mell et al., 2012). Mell et al. (Mell et al., 2012) showed that many pairwise interactions contribute to uptake 5-fold more than the expected contribution of individual bases alone. This suggests that uptake is determined not only by the contribution of individual bases but also by interactions between them. Main questions about uptake bias and self-specificity Self-specificity, defined as the preference of bacteria for taking up DNA from its own or closely related species,  has been seen as a mate-recognition mechanism to avoid negative consequences of DNA recombination (such as inefficient codon usage and loss of coadapted alleles) (Baltrus, 2013). However, it is better explained by genomic enrichment of sequences that bind more effectively to the receptor. Some degree of uptake bias is expected to be universal in all competent species. When the conditions for uptake bias are met, molecular drive will lead to self-specificity. 11 My study aims at increasing the understanding of uptake bias by 1) determining the factors explaining the differences in DNA uptake across the genome of a species with self-specificity and 2) testing for uptake biases in a species without self-specificity.  Work in chapter 2 used H. influenzae as a model to understand factors influencing DNA uptake in species with self-specificity.  H. influenzae was chosen given its strong uptake bias and self-specificity. The goal of this study was to measure DNA uptake at every position in the H. influenzae genome and to use these data to characterize the DNA uptake biases caused by the USS and any other sequence factors. To elucidate which factors, have a greater influence on DNA uptake, I first developed a naive computational model that predicted the effect of uptake sequences on DNA uptake across the H. influenzae genome using the distribution of USS in the genome. This model was used as a framework to which changes could be made based on what we learned from the actual measurements of DNA uptake produced by sequencing genomic DNA fragments that had been recovered after being taken up by competent H. influenzae cells. The role of the model was not to use the observed uptake data to build the most accurate model that will include all factors relevant to uptake. This is because the molecular steps involved in DNA binding and uptake are only partially understood, so the steps simulated by the model must necessarily include assumptions about these steps since they cannot be addressed by our uptake experiments. Instead, the role of the model was to use the discrepancies between predicted and observed uptake as a tool to reveal the strength of the bias, effects of USS sequence differences, and the influence of the distribution of USS locations. These factors, in turn, increased the understanding of the genomic distribution of recombination and the effects of competition with DNA from the host or other microbiota. The work described in Chapter 3 tested whether uptake biases are present in a species that showed no preferences in taking up genomic DNA from different species (without self-specificity). A. baylyi was chosen as a model, given it lacks self-specificity and it has high levels of natural transformation (Palmen et al., 1993, 1994). This was done by first having competent A. baylyi take up a 200 bp synthetic fragments with a 69 bp fully randomized segment, then recovering and sequencing the DNA that had been taken up.  Finding that the taken-up DNA contained overrepresented sequences would mean that A. baylyi has uptake biased for specific 12 sequences, supporting the hypothesis that uptake biases are a generalized property of DNA uptake. Since A. baylyi lacks self-specificity, the preferred motif is expected to be short enough to be in mildly enriched in its genome. However, no sequences were overrepresented in the recovered fragments. Absence of uptake preferences in A. baylyi, suggest that such biases are not required to increase the strength of the physical interactions between proteins of the uptake machinery and DNA. This imply that uptake biases are not universal. However, this is not contradictory with self-specificity evolving as a consequence of molecular drive when strong uptake biases are present.    13  CHAPTER TWO Genome-wide analysis of DNA uptake across the outer membrane of naturally competent Haemophilus influenzae  Introduction Many bacteria are naturally competent, able to actively bind DNA fragments at the cell surface and pull them into the cytoplasm, where the incoming fragments may contribute nucleotides to cellular pools or recombine with homologous genomic sequences (Lorenz and Wackernagel, 1994). The genetic exchange associated with this latter process contributes to adaptation and is known to have promoted resistance to antibiotics (Bae et al., 2014) and increased strains’ intracellular invasiveness (Mell et al., 2016) and vaccine resistance (Kress-Bennett et al., 2016; Straume et al., 2015). Thus, understanding how different genomic regions evolve via natural transformation processes could be used to predict the spread of pathogenic traits. Most competent bacteria that have been tested take up DNA regardless of sequence, but species in two families, the Pasteurellaceae and the Neisseriaceae, exhibit strong preferences for DNA containing short sequence motifs (Chen and Dubnau, 2004). Because these motifs have become highly enriched in the corresponding genomes, these biases effectively limit uptake to DNA from close relatives with the same uptake specificity (Dougherty et al., 1979; Scocca et al., 1974). The distribution of the preferred sequences around the chromosome is uneven (Smith et al., 1995), which may cause different genes to experience quite different rates of genetic exchange.  14 Most steps in the natural transformation process are highly conserved among transformable species (Chen and Dubnau, 2004). In the Pasteurellaceae, the Neisseriaceae and most other Gram-negative bacteria, DNA uptake is initiated by binding of a type IV pilus uptake machine to double-stranded DNA (dsDNA) at the cell surface. The DNA-binding protein has not been identified in H. influenzae, but in Neisseria it is a minor pilin-type protein that forms part of the pilus (Cehovin et al., 2013). DNA binding is followed by retraction of the pilus, which pulls the DNA across the outer membrane into the periplasm. Because circular DNA is taken up as efficiently as linear DNA, uptake is thought to begin internally on DNA fragments rather than at their ends (Barany et al., 1983). Thus, it is likely that the stiff dsDNA molecule is transiently kinked (folded sharply back on itself) at the site of initiation to allow it to pass through the narrow secretin pore of the uptake machinery. Forces generated by the retraction of the type IV pilus are thought to be responsible for this kinking, which might be facilitated by strand separation at the AT-tracts (Danner et al., 1982). Once a loop of the DNA is inside the periplasm, a ratchet process controlled by the periplasmic protein ComEA is thought to pull the rest of the DNA through the outer membrane (Hepp and Maier, 2016; Salzer et al., 2016). Below we use ‘DNA uptake’ to refer to the combined binding and membrane-transport steps that move DNA from the extracellular environment across the outer membrane into the periplasm. In Gram-positive bacteria similar machinery acts to pull the DNA through the thick cell wall (Chen and Dubnau, 2004). The next step in natural transformation is translocation of the DNA out of the periplasm and into the cytoplasm. Only the 3’-leading strand remains intact, passing through an inner membrane pore encoded by the rec2/comEC gene, while the other strand is degraded in the periplasm and its nucleotides are dephosphorylated and imported as nucleosides (Pifer and Smith, 1985). Circular DNA molecules are efficiently taken up across the outer membrane but remain the periplasm because they lack free ends (Pifer and Smith, 1985). As the single strand enters the cytoplasm it undergoes limited exonucleolytic degradation before being complexed with cellular proteins. If sequence similarity permits the strand may then recombine with homologous chromosomal sequences; otherwise the strand is degraded to its constituent nucleotides (de Vries et al., 2001). Both linkage and sequencing studies indicate that fragments much longer than the cell are readily taken up  (Goodgal, 1982; Mell and Redfield, 2014), as are 15 fragments as short as 200 bp, although a lower limit has not been established (Maughan and Redfield, 2009; Mell et al., 2012). However, recombination of short fragments is limited because they are usually degraded by cytoplasmic nucleases before they can recombine (Pifer and Smith, 1985). Uptake speed has been estimated at 500-1000 bp/sec, with transformation essentially complete by 15 min (Deich and Smith, 1980). Direct measures of DNA uptake bias: Uptake-competition experiments in the Pasteurellacean Haemophilus influenzae and in Neisseria gonorrhoeae showed that genetically marked ‘self-derived’ DNA competes for uptake with unmarked self-derived DNA but not with DNA from unrelated sources (Dougherty et al., 1979; Scocca et al., 1974). Subsequent DNA uptake experiments using cloned radiolabeled DNA fragments found that these self-preferences are caused by the uptake machineries’ strong biases for short sequence motifs, called uptake signal sequences (USS) in H. influenzae and DNA uptake sequences (DUS) in Neisseria species (Sisco and Smith, 1979) (Davidsen et al., 2004). Sequence comparisons and site-directed mutagenesis initially identified the H. influenzae motif as an 11 bp sequence with a strong contribution by flanking AT-rich sequences (Danner et al., 1980, 1982), and later genome sequencing identified 1465 occurrences of a 9 bp USS core in H. influenzae and 1892 occurrences of an unrelated 10 bp DUS motif in N. meningitidis (Smith et al., 1995, 1999). Subsequent motif search analyses by Maughan et al. (Maughan et al., 2010) expanded this number to 2206 USSs in the genome of the standard H. influenzae lab strain Rd, sharing the motif shown in Figure 2.1A. These genomic analyses were later complemented by direct uptake experiments using mutated and degenerate USS variants (Maughan et al., 2010; Mell et al., 2012). Mell & collaborators (19) used mutagenesis and sequencing of pools of degenerate USS-containing fragments that had been recovered after uptake to identify the contribution to uptake of each USS position, and found that the central GCGG bases are crucial for uptake, with much smaller contributions made by the flanking bases and AT-rich segments. The motif in Figure 2.1B shows the contribution of each base considered independently. Interaction effects between bases of the AT-tracts and the core were also found to make important contributions to uptake, but only a few of these have been directly measured. Although characterization of the unrelated Neisseriacean DUS has not reached this level of detail (Mathis and Scocca, 1982), the 16 two uptake systems shares many features, apparently convergently evolved, including the presence within each family of lineages with slightly different preferred motifs (Frye et al., 2013; Redfield et al., 2006).   Figure 2.1. The H. influenzae uptake signal sequence. A. Sequence logo showing the individual contributions to genomic abundance of bases in the USS motif (Maughan 2010). B. Sequence logo showing the individual contributions to uptake of bases in the USS motif, as measured by Mell et al. (2012). C. Conserved USS segments.   Evolution of uptake sequences in the genome: Alignment of homologous genomic regions from different Pasteurellaceae species showed that USS evolve by point mutations (Redfield et al., 2006); i.e. they are not inserted elements. As a mechanism for this evolution, (Danner et al., 1980) proposed that the combination of uptake bias and genomic recombination CoreInner core AT-richsegmentAT-richsegmentC.B.A.2 4 306 8 10 12 14 16 18 20 22 24 26 28Positions12Bits012Bits017 creates an evolutionary pressure that causes the preferred uptake sequences to accumulate throughout the genome, with their numbers limited by their eventual interference with gene function. Consistent with this, uptake sequences in both Haemophilus and Neisseria are underrepresented in newly acquired segments, in rRNA genes, and in coding sequences, especially those with strong functional constraints (Findlay and Redfield, 2009; Smith et al., 1999). Modeling by Maughan et al. (Maughan et al., 2010) confirmed that this molecular drive process could produce uptake sequence distributions like those of real genomes, with no need for any fitness benefit from either the uptake sequences or the recombination they promote. Thus, the presence of biased DNA uptake machinery may be sufficient in itself to explain the abundance of uptake sequences. Such biases may be solely a consequence of direct selection on the DNA uptake machinery for more effective DNA binding or may have been reinforced by indirect selection for preferential uptake of conspecific DNA. These processes might be especially important in respiratory tracts and other mucosal environments where Pasteurellaceae and Neisseriaceae species mainly occur (Man et al., 2017). These environments contain abundant host DNA, and transformation can only occur if the released bacterial DNA competes successfully for binding to the uptake machinery (Lethem et al., 1990; Shak et al., 1990).  The goal of the present study was to measure DNA uptake at every position in the H. influenzae genome and to use this uptake data to characterize the effects of USS and identify any other factors affecting uptake. I measured uptake by sequencing H. influenzae genomic DNA, pre-sheared into short (50-800 bp) and long (1.5-17 kb) sizes, from strains 86-028NP and PittGG that has been taken up by the H. influenzae strain Rd. To prepare a framework for interpreting uptake biases, we developed a computational model that predicted the effect of uptake sequences on DNA uptake across the H. influenzae genome. We did not attempt to build a model that accurately simulated the actual events of DNA uptake, since too little is known about these. Instead, the initial version of this model was ‘naïve’ in that its parameters and settings were based only on previously published information. Discrepancies between the model’s predictions and the observed uptake were then used to identify features of uptake that were poorly predicted. Hypothesized biological explanations for these discrepancies then guided changes to the model, and the effect of each change on the discrepancy was used to confirm or refute the 18 hypothesis. The most important product of this recursive analysis was not the model itself, but the improved understanding of factors affecting DNA uptake across the genome. These in turn increased the understanding of the genomic distribution of recombination and the effects of competition with DNA from the host or other microbiota. Methods Identifying USSs in the genomes. Genomic USSs were identified by scoring each genome position with the position-specific scoring matrix (PSSM) of Mell et al. (2012); this is based on uptake of synthetic fragments containing degenerate USS sequences. Positions scoring ≥ 10.0 or ≥ 9.5 (maximum score is 12.6) were included in the standard (USS10 and USS9.5) lists of USS locations. Since USS are asymmetric, USS positions in both orientations were specified by the location of their central base 16. Sequence logos of USSs were generated using R package seqLogo v. 3.8.  Predicting DNA uptake from DNA sequence. The predictive model was written in R v.3.5.1. Given a list of USS positions and scores in a DNA genome of specified length, it used a specified distribution of DNA fragment lengths or length bins (e.g. 1-100 bp, 101-200 bp, etc.) to calculate the relative uptake of every position in a circular genome. At each DNA position in turn, for each fragment length or bin, the model summed the predicted uptake contributions for every fragment of that length that overlapped the position.  For efficiency, the full calculation was only done for the first position. At each subsequent position, the model calculated the new sum from the previous position’s sum by subtracting the contribution of the formerly leftmost fragment and adding the contribution of the new rightmost fragment (Figure 2.3A).   Each fragment’s predicted contribution to uptake depended on the number of USS it contained, and on the scores and relative locations of these USS. Fragments with no or incomplete USSs were assigned baseline values for the probabilities of being bound (p_bind) and taken up (p_uptake); initial values for both were arbitrarily set to 0.1.  For fragments with one or more complete USS10, p_bind was calculated as 1 – mean_gap/20000, where mean_gap was the mean length of USS-free segments in the fragment and 20000 the maximum fragment length in bp. The uptake function p_uptake was initially specified as p_uptake = 0.1 + (1 - 0.1)/(1 + exp(-5 * (score – 11))).  19 Once the model had calculated the contributions of a specific fragment length or length bin to uptake of every genome position, it moved on to the next length or bin. Once the contributions of every length or bin had been calculated, the model combined all the contributions for each position, taking into account the frequency of each length or bin in the input DNA. These position-specific uptake predictions were then normalized to a mean genome-wide uptake value of 1.0. In response to ongoing analysis of the 86-028NP DNA uptake data, the initial model underwent modifications to improve its predictions for a ‘far from USS10’ subset of positions that were at least 0.5kb from a USS9.5. (n = 361965 positions), combined with 209 USS9.5 peak positions separated from the nearest USS10s by at least 1000b bp. This reduced the baseline p_uptake of USS-free fragments from 0.1 to 0.005 and the USS cutoff score from 10 to 9.5, excluded from consideration USS9.5 that were within 50 bp of fragment ends, and identified better slope and inflection point values for the sigmoidal uptake function using the R function “nls” from the stats-package.  These changes replaced the previous uptake function with p_uptake = 0.005 + (1 - 0.005)/(1 + exp(-3.8 * (score – 10.6))). A subsequent change adjusted uptake predictions according to the GC content around each position. First the observed effect of GC content on uptake was approximated by a linear function describing how the 86-028NP long fragment uptake ratios depended on their local GC contents calculated with a 2 kb window (see inset in Fig. GC).  Each genomic position was assigned the GC content of a 1001 bp window centered on it. The predicted uptake at each position was then modified using the local GC content and the function.  Bacterial strains, culturing, and competent cell preparations: The KW20 recipient strains were rec2 derivatives of the standard H. influenzae lab strain Rd KW20, with (RR3117) and without (RR3125) a spectinomycin resistance allele (Mell et al., 2012; Sinha et al., 2012). The 86-028NP donor strain (RR3133) was a derivative with a nalidixic acid resistance allele (Mell et al., 2011); the PittGG isolate was unmodified. Standard growth and culturing methods were used (Poje and Redfield, 2003) (Table 2.1); liquid cultures were grown with shaking at 37 °C in brain-heart infusion broth supplemented with NAD (2µg/ml) and hemin (10 µg/ml) (sBHI), with 1.2% agar added for plate cultures. To prepare naturally competent cells, cultures were first maintained in 20 exponential growth at OD600 below 0.2 for at least 2 hr, and at OD600 = 0.2 cells were collected by filtration from 10 ml of culture, transferred into 10 ml of starvation medium M-IV, and incubated at 37 °C for 100 minutes before DNA uptake experiments (Poje and Redfield, 2003).  Table	2.1.	Haemophilus	influenzae	strains	used	in	this	study	Strain number Strain name  Phenotype Source RR3117 rec2::spec Spectinomycin resistant Rd derivative. No translocation of taken-up fragments to the cytoplasm Sinha et al. 2012 RR3125 Rd 𝚫rec2 Unmarked Rd derivative. No translocation of taken-up fragments to the cytoplasm Sinha et al. 2012 RR3133 86-028NP NalR Otitis media clinical isolate. Nalidixic acid resistant Mell et al. 2011 RR1361 PittGG Nontypeable clinical isolate G. Ehrlich RR722 Rd Rd KW20, rough (unencapsulated) derivative of type d H. O. Smith  Input DNA preparations. High molecular weight donor DNA was purified using standard phenol:chloroform extractions (Sambrook, 2001) from 10 ml overnight cultures of the 86-028NP derivative, and PittGG carrying selectable markers (Table 1). This DNA was then sheared into separate ‘long fragment’ (1.5-9 kb) and ‘short fragment’ (50-500 bp) preparations using Covaris G-tubes and sonication respectively. The fragment length distributions were measured using a 21 Bioanalyzer with a DNA 12000 kit (Agilent), dividing the relative fluorescence at each time point by its fragment length estimated from the size standards.  DNA uptake and recovery. 10 ml of competent rec-2 mutant Rd cells in MIV were incubated with 10 µg of sheared donor DNA for 20 min at 37 °C. To degrade remaining free DNA, the culture was incubated with 1 ug/ml of DNase I for 5 minutes. Cells were washed twice by pelleting and resuspension in cold MIV, and the final pellet was rinsed twice with cold MIV before resuspension in 0.5 ml of extraction buffer (Tris-HCL 10 mM pH 7.5, EDTA 10 mM, CsCl 1.0 M). Periplasmic DNA was extracted using the organic phenol:acetone extraction method as described by Mell et al. 2012 (Barouki and Smith, 1985; Kahn et al., 1983; Mell et al., 2012) followed by an ethanol precipitation. DNA was resuspended in 20 µl of T10E10 buffer (Tris-HCl 10 mM pH 7.5, EDTA 10 mM). The DNA was then incubated at 37 °C with 400 ng of RNase A for 1 hour, followed by 30 min incubation with 30 ng of proteinase K to remove RNase A. Recovered DNA was then separated from longer fragments of contaminating genomic DNA by electrophoresis in a 0.8% agarose gel and recovered from the gel slice with a Zymo gel DNA recovery kit. Recovered periplasmic DNA was quantified using a Qubit dsDNA HS Assay Kit (absolute DNA concentration).  DNA sequencing and data processing. Sequencing libraries of the input and taken-up DNA samples were prepared using Illumina Nextera XT DNA library prep kits according to manufacturer recommendations. An Illumina NextSeq500 was used to generate 1-10 million paired-end reads of 2x150 nt for each library (giving >100-fold genomic coverage). Summary statistics for each sample are provided in Table S1 (see Appendix).  Reference sequences: The original PittGG reference (NC_009567.1) generated by pyrosequencing had many indel errors, so a new reference was constructed by Pacific Biosciences RSII of our laboratory version of this strain (RR1361) (assembly by HGAP2 v2.0, followed by Circlator (Hunt et al., 2015), and then Quiver to polish the circular junction). For our analysis the NCBI sequence references for PittGG, 86-028NP, (NC_007146.2) and Rd KW20 (NC_000907.1) were then further corrected using Pilon v1.22 and the new Illumina reads of input or control samples. This was particularly important for the Rd KW20 recipient reference, since the original sequence dates from 1995 (Fleischmann et al., 1995) and contains several 22 hundred ambiguous bases and errors. This correction step also accommodated the presence of the nalidixic acid resistance marker in 86-028NP. Competition essays simulations of H. influenzae with human DNA, used 3 random segments of the same size as H. influenzae 86-028NP genome (1914386 bp) the Chromosome 1 (CM000663.2, positions 33610150 – 35524535), Chromosome 3 (CM000665.2, positions 1348752 – 3263137), and Chromosome 12 (CM000674.2, positions 18170588 – 20084973). Respiratory bacterial genomes used to score USS10 and USS11 in table 2 and competitions essays were Streptococcus pneumoniae R6 (NC_003098.1), Neisseria meningitidis MC58 (NC_003112.2), Pseudomonas aeruginosa PAO1 (NC_002516.2), Aggregatibacter actinomycetemcomitans VT1169 (NZ_CP012958.1), Haemophilus parainfluenzae T3T1 (NC_015964.1), Haemophilus ducreyi 35000HP (NC_002940.2), Mannheimia haemolytica M42548 (NC_021082.1). Chromosomal contamination measurements  and corrections: Reads from the recipient genomic DNA that contaminated taken-up DNA samples were identified by a ‘competitive alignment’ step that aligned all the sample’s reads (using bwa mem v0.7.15, samblaster v0.1.24, and sambamba v0.5.0) to a concatenated double-reference sequence consisting of the recipient Rd genome and the donor genome (86-028NP or PittGG). Because the donor and recipient genomes are distinguished by a high density of SNVs, as well as structural variation and large indels (Harrison et al., 2005; Hogg et al., 2007; Mell et al., 2011), most contaminating Rd reads in uptake samples aligned only to the Rd reference while most of the desired donor-derived reads aligned only to the donor reference. Reads that mapped equally well to both genomes or to repetitive sequences within a genome were identified by their MAPQ scores of 0 and were removed from the analysis. The numbers of reads mapping uniquely to either donor or recipient genome were then used to calculate the contamination level of each taken-up DNA sample, as the ratio of recipient-mapping reads to total uniquely mapping reads (Table S1).  Subsequent depth of coverage values and summary statistics were extracted for all positions or specific intervals using bedtools coverage v2.16.2 or sambamba flagstat (Table S1). All subsequent analyses and plotting used the R statistical programming language, including standard add-on packages dplyr, tidyr, plyr, ggplot2, data.table. Other packages used are specified below. Code is available at https://github.com/mamora/DNA_uptake.  23 Calculation of experimental uptake ratios from sequence coverage. After contaminating reads had been removed from each sample, uptake maps for each donor DNA were created by dividing the mean of the three normalized taken-up-DNA coverages for each position by the corresponding normalized input-DNA coverage. Finally, uptake ratios were normalized to a genome-wide mean uptake of 1.0 and smoothed by calculating the mean uptake over a 31 bp central-oriented sliding window using function rollapply from R package zoo v. 1.8-5. The effects of this smoothing are shown for the peak examples in Figure S5. Periodicity analysis:  To detect possible periodic patterns in coverage depth and in uptake ratios for the four datasets, periodograms were created using the R package TSA v. 1.2. Analysis of uptake ratio data:  To obtain a set of well-isolated USS10s for analysis of peak shapes, we identified the closest peak separation at which USS effects did not overlap by examining sets of USS10 that were separated by different distances (1200, 1000, 800, 600 bp), excluding positions with missing data and USS10 that were 400 bp or less from positions with low input coverage (≤ 20 reads). Separation of ≥1000 bp was found to give the best compromise between good peak separation and the number of USS meeting the separation criterion (237 USS10s and 209 USS9.5s).  The search for non-USS sequences causing weak uptake effects used a subset of positions that were at least 0.6kb from the closest USS10. This gave 575 ‘far from USS’ segments summing to 29% of the genome. Uptake maps of segments containing positions with uptake ratios > 0.2 were examined visually to distinguish between (i) shoulders of adjacent USS10 peaks, (ii) increased uptake at USS with scores between 9.5 and 10, and (iii) increased uptake at non-USS sequences.  Incorporating within-USS interaction effects into uptake predictions:  Figure 2.14 of Mell et al. (Mell et al., 2012) shows the strength and direction of pairwise interaction effects between positions on the same USS , inferred from uptake analysis of synthetic degenerate USS. From this figure we extracted the mid-range value of the interaction effect at each interacting pair of USS positions (only some pairs of positions showed such effects). For each 86-028NP USS9.5 whose sequence differed from the USS consensus at both positions of such a pair, the USS 24 score was modified by adding or subtracting the corresponding interaction value. The modified scores were then used by the model to predict DNA uptake, as described above.  Simulated noise analysis: Simulated noise-free uptake data for short and long fragments was first generated by smoothing raw uptake coverage data for 86-028NP-short (sample UP7) and 86-028NP-long (sample UP3) using a LOESS regression, and normalizing the results to a mean coverage of 1.0. Simulated relative-noise amplitudes for every genome position were generated using the ‘tuneR’ R-package (Ligges et al., 2018). Before being added to the noise-free data, the noise amplitude at each position was adjusted in proportion to the noise-free simulated coverage at that position, with the maximum noise range for each coverage level set by a multiplier (1.0, 1.5, 2.0, 2.5 or 3.0) and by the range of all experimental replicates for positions with that mean coverage.  Red noise was used because, when added to the simulated noise-free coverage it gave an autocorrelation of 0.999, identical to that of the experimental data. Other noise types were evaluated but not used, since their autocorrelations were lower (0.975 for pink noise and 0.836 for white noise).  Results A computational model of DNA uptake As a framework for interpreting DNA uptake data we developed a simulation model of USS-dependent DNA uptake. It takes as input the locations and strengths of USSs in the DNA whose uptake is to be simulated, the fragment-size distribution of this DNA, and binding and uptake functions that specify how uptake probability depends on USS presence and strength. The output is the expected relative uptake of every position in the genome. Development of the model was guided by basic principles of sequence-specific protein-DNA interactions (Halford and Marko, 2004; Rohs et al., 2010). The first step in these interactions is thought to be a random encounter between a DNA fragment and the binding site of the protein, usually at a DNA position that does not contain the protein’s preferred sequence. This non-specific binding dramatically increases the probability that the protein will subsequently encounter any preferred sequence, either by sliding along the DNA or by transient dissociation 25 and reassociation, leading to specific binding between DNA and protein. In the case of the USS this specific binding then enables uptake of the DNA fragment across the cell’s outer membrane.  The model did not explicitly simulate the first step, non-specific binding, since this is expected to be equally probable for all DNA positions. The specific binding and DNA uptake steps were separately modeled since they are expected to depend on the properties of the DNA uptake machinery and on the length and sequence of the DNA fragment. Although in real cells both steps may depend on the quality of the USS, for simplicity the initial version of the model assumed that specific binding required only a threshold similarity to the USS consensus and that the subsequent probability of uptake depended on the strength of this similarity. Simulating these steps required first specifying the genomic sequences that should be treated as USS. This was not straightforward because genomes contain many USS variants that differ in how well they promote DNA uptake (Findlay and Redfield, 2009; Maughan and Redfield, 2009). Our strategy was to score the information content (in bits) of every genome position, using the Position-Specific Scoring Matrix (PSSM) from Mell et al.’s degenerate-sequence uptake experiment (Mell et al., 2012) (Table S2 (see Appendix)), and to use overrepresentation of high-scoring sequences as the USS criterion. We scored every genome position in the three H. influenzae strains used for the experiments described below (Table 2.1), and in four randomly generated sequences with the same length and base composition (Figure 2.2). In the H. influenzae genomes, overrepresentation of high-scoring sequences was detectable above a score of 7.0 bits and became dramatic above 10.0 bits, where the numbers of high-scoring positions increased in H. influenzae genomes but became vanishingly small in the random-sequence controls, (see inset in Figure 2.2). Since the slight overrepresentation of scores between 7 and 10 bits was hypothesized to be not a direct effect of DNA uptake but an indirect consequence of mutational degeneration of high scoring USSs, the model initially used a USS cut-off score of 10 bits (‘USS10’, n=1941 in strain 86-028NP). This was later reduced to a less stringent 9.5 bits after weak uptake effects had been examined (’USS9.5’, n=2248 in strain 86-028NP). 26  Figure 2.2. Frequency distribution of USS scores for all positions in H. influenzae and random-sequence genomes. Related to Figure 1. 86-028 NP (blue), PittGG (orange), Rd (gold), and four random-sequence 1.9 Mb genomes with the same base composition (38%G+C, black and grey).  Scores were calculated with the uptake scoring matrix in Table S2. The numbers in the lower right are the numbers of positions meeting cutoff scores of 9.5, 10.0 and 10.5 bits. Inset: Expanded view for positions with scores higher than 9 bits.  The computational model used these USS scores to predict DNA uptake for every position in the genome, summing the contributions of binding and uptake probabilities from DNA fragments of different sizes (Figure 2.3A gives an overview). Each fragment under consideration was first checked for the locations of any USS10s, and the maximum fragment length (20 kb) and the mean length of USS-free segments (mean_gap) were used to calculate the probability that a DNA receptor protein initially encountering a random location in the fragment would then encounter and bind specifically to a USS10 rather than disassociating from the fragment: p_bind = 1 – mean_gap/20000). Fragments with no USS were initially assigned a baseline p_bind of 0.1. The probability that this specific binding led to uptake of the DNA fragment was then calculated from the USS10 score (or the mean score if the fragment contained more than one USS10), using the function p_uptake = 0.1 + (1 - 0.1)/(1 + exp(-5 * (score – 11))), where 0.1 specifies the baseline uptake of USS-free fragments and -5 is an arbitrarily chosen coefficient specifying the FrequencyFrequencyUSS score (bits)1941/1929/19012248/2227/21781734/1729/17020.01 -0.00 -0.02 -0.03 -9 10 11 124e-5 2e-5 -0.0 ---- - --RdRandom seq.PittGG86-028NP27 slope at the inflection point. The score of 11 bits specifying the inflection point of the function was chosen because this score was the midpoint of the range of increasing USS overabundance in the genomes (see Figure 2.2). When combined with the baseline p_bind of 0.1, the p_uptake baseline assigned fragments with no USS10 a net uptake probability of 0.01.  Once the contributions of every size of fragment had been calculated for each position (Figure 2.3A), the model combined all the contributions, taking into account the frequency of each size in the input DNA. The position-specific uptake predictions were then normalized to a genome-wide mean uptake probability of 1.0. Except in simple test cases, for computational efficiency DNA fragment lengths were specified as the median lengths of bins (10 bp bins for short fragments, 200 bp bins for long fragments) rather than each being considered separately (e.g. 101-120 bp, 121-140 bp). This initial model does not use data from the uptake experiments.  The strategy was to begin with a model that incorporated what was known before the uptake experiment was analyzed, and then refine it with information from our analysis. Initial p_bind was estimated based on what it is known about proteins binding to DNA.  DNA binding proteins most likely initially bind without specificity to a random place of a DNA molecule and then slide through fragments scanning for target binding sequence. The mean_gap argument measured the average length of spaces in the fragments without a USS. This calculation allowed me to estimate the probability of the DNA-receptor protein to find a USS before detaching from a fragment. On the other hand, initial p_uptake was estimated using what it is known about USS distribution in the genome, as explained above, using an inflection point based on the midpoint of the range of increasing USS overabundance in the genomes (see Figure 2.2).  Model results: Figures 2.3B and 2.3C show examples of model predictions for simple situations. Figure 2.3B shows the uptake predictions for an 800-bp simulated genome containing a single USS with score 12.0 bits, considering three different input DNA fragment sizes (100, 200 and 300 bp). The peaks at the USS have straight sides, a basal width twice the length of the fragments being taken up, and 31-bp flat tops arising from the model’s requirement for a full-length USS. When the DNA fragment sizes were evenly distributed between 25-300 bp in length 28 (Figure 2C), the peak had steep sides at its tops and gradually flattened at the base; maximum width at the base equalled twice the maximum fragment length. In simulated mini-genomes with more than one USS (Figures 2.3D, E and F), isolated peaks were only seen when the DNA fragments being taken up were substantially shorter than the spacing of the USSs (Figure 2D), and peaks disappeared entirely when the fragments were long enough that almost all contained at least one USS (Figure 2.3F).  Figure 2.3G and H show the predicted uptake maps when this model analyzed a 50kb segment of the H. influenzae 86-028NP genome, using the short-fragment and long-fragment length distributions from the actual uptake experiments described below (Figures 2.4A and B), and Figure 2.3.I shows the distribution of USSs over this segment. Because the ‘short’ DNA fragments are shorter than the typical separation between USSs, the model predicts that uptake will be restricted to sharp peaks at each USS. In contrast, uptake of long DNA fragments is predicted to be much more uniform, since most of these will contain at least one USS. 29  Figure 2.3. A computational model to predict DNA uptake. A. Components of the DNA uptake model (see Methods and Results for details). B. & C. Model predictions for uptake centered at a 12-bit USS for: B. 100, 200, and 300 bp fragments, C. a mixed distribution of fragments between 25-300 bp with and without baseline uptake. D. E. & F. Model predictions for uptake of a 3000 USSFocal genome positionSection of the genome USSFragments with no USS Fragments of this size contributing tofocal  positionuptake Fragments with 1 USS Fragments with 2 USS Former ‘first’ fragmentNext ‘last’ fragmentFragment size binsFrequencyA.100 bp200 bp300 bpMixed fragment sizes0Position relative to USS  (bp)-200 200 0-200 2008462Relative uptakeBaseline uptakeNo baselineuptake  400 1200 2000 2800Position (bp)25-300 bp fragments50-2000 bp fragments1- 14 kb fragmentsRelative uptake10 11 1202419.012.6USS10scorePredicted Uptake ratio10000 20000 30000 40000 50000086-028NP genome position (bp)F.H.I.E.D.C.B.G.30 bp region with 3 USSs (red squares, scores in black) using different fragment-length distributions: D. 50-300 bp fragments, E. 50-2000 bp fragment, F. 1-14 kb fragments. G. & H. Predicted DNA uptake of a 50 kb segment of the 86-028NP genome using G. short and H. long fragment-length distributions: I. Locations and scores of USS10s in this 50 kb segment.     Figure 2.4: Distributions of fragment lengths in input DNA preparations. Relative abundances of DNA fragment lengths were estimated from Bioanalyzer data for input DNA samples from strains 86-028NP (blue) and PittGG (orange).  A. Short-fragment preparations. B. Long-fragment preparations. Solid lines: length distributions of fragments in input DNA preparations, normalized to most frequent length.  Dashed lines: Length distributions of sequenced fragments, with arbitrary scaling. Insets: Bioanalyzer pseudo-gel images of sheared DNAs (NP: 86-028NP, GG: PittGG). Bioanalyzer molecular weight markers are shown in purple and green.   Generation of experimental DNA uptake data To obtain high-resolution measurements of actual DNA uptake, we sequenced H. influenzae genomic DNA that had been taken up by and recovered from competent H. influenzae cells. Competent cells of the standard laboratory strain Rd were first incubated with genomic DNA Relative fragment number00.20.40.60.811.20 2000 4000 6000 8000 10000 12000 14000 16000 18000UP13 norm to maxFragment size (kb)1.004 8 12 16000.20.40.60.811.20 200 400 600 800 1000UP15 Normalized # frags1.000.2 0.4 0.6 0.8 .A.B. NP  GGNP  GG- 17 kb- 50 bp- 17 kb- 50 bpPittGG86-028NP31 preparations from strains 86-028NP and PittGG, whose core genomes are readily differentiated from Rd (and each other) because they differ at ~3% of orthologous positions (Hogg et al., 2007). To allow efficient recovery of the taken-up DNA, the Rd strain in which competence was induced carried a rec2 mutation that blocks translocation of taken-up DNA fragments, causing the DNA to be trapped intact in the periplasm. The 86-028NP and PittGG genomic DNAs were pre-sheared to give short (50-800 bp) and long (1.5-17 kb) DNA preparations (size distributions are shown in Figure 2.4), and three replicate uptake experiments were done with each DNA preparation. After 20 min incubation with competent cells, the taken-up DNA was recovered from the cell periplasm using the cell-fractionation procedure of Kahn et al. (Barouki and Smith, 1985; Kahn et al., 1983; Mell et al., 2012). Taken-up DNA samples were sequenced along with samples of the input 86-028NP and PittGG DNAs and of the recipient Rd DNA. The input and uptake reads were then aligned to the corresponding 86-028NP and PittGG reference sequences and coverage at every position was calculated. Table S1 (see Appendix) provides detailed information about the four input samples, the twelve uptake samples, and the Rd sample.  Removal of contaminating Rd DNA: Preparations of DNA recovered from the periplasm after uptake always included some contaminating DNA from the recipient Rd chromosome. The divergence between the Rd and donor genomes allowed us to estimate the extent of this contamination by competitively aligning the taken-up reads from each sample to an artificial reference ‘genome’ consisting of both recipient and donor genomes as separate ‘chromosomes’. Reads that uniquely aligned to only one chromosome could then be unambiguously assigned to either donor (taken-up) or recipient (contamination). The resulting estimates of Rd chromosomal contamination were between 3.2% and 19.3% of reads; sample-specific values are listed in Table S1 (see Appendix).  The effects of this contamination were not expected to be uniform across each donor genome, since segments of the 86-028NP and PittGG genomes with high divergence from or with no close homolog in Rd would be free of contamination-derived reads. We used the competitive-alignment described above to create contamination-corrected uptake coverages by discarding all reads that could not be uniquely mapped to the donor genome; in addition to removing Rd contamination this also removed reads from segments that are identical between the donor and 32 recipient strains (‘double-mapping reads’) and reads that mapped to repeats, such as the six copies of the rRNA genes. For consistency, the same changes were applied to the input samples although they did not experience any contamination. This correction removed an average of 18.6% of reads (range 8.9%-28.3%), left some segments of the 86-028NP and PittGG genomes with no coverage in all samples (2.3% and 2.1% respectively), and reduced coverage adjacent to these segments (Figure 2.5). Contamination details for each sample are provided in Table S1, and the impacts are considered below. The uptake analysis described below showed this correction to be effective.   Figure 2.5 Locations of positions with missing uptake ratio data. Each point represents a genome position for which an uptake ratio could not be calculated. The points are vertically jittered, so segments with no coverage appear as black rectangles.  A. 86-028NP, short-fragment data.  B. 86-028NP, long-fragment data. C. PittGG, short-fragment data. D. PittGG, long-fragment data. Uptake ratios: To control for position-specific differences in sequencing efficiency, contamination-corrected read coverage at each position in each uptake sample was divided by read coverage at that position in the corresponding input sample (e.g. coverages of each 86-028NP-short uptake sample were divided by 86-028NP-short input coverages). Normalizing the mean of the three replicates to a genome-wide mean ratio of 1.0 then gave a mean ‘uptake ratio’ measurement for each genome position for each DNA type. Finally, each position’s uptake C.|0|500,000|1,000,000|1,500,000D.Positions with missing uptake ratio data (bp)|A.| | | |B.33 ratio was smoothed using a USS-length (31 bp) window. Figures 2.5 and 2.6 show the resulting uptake ratio maps. Figure 2.6A shows the short-fragment uptake ratio map for the first 50kb of the 86-028NP genome; the ticks in Figure 2.6C indicate locations and scores of USS10s. The pattern is strikingly similar to that predicted for the same DNA segment by the model (Figure 2.3G). Sharp uptake peaks are seen at USS10 positions, some separated by flat-bottomed valleys and others overlapping. Figure S2.7A, B and C show similar analysis for the first 50 kb of strain PittGG’s genome, and Figure 2.8A and B show expanded maps for two examples of the 86-028NP peaks. The full-genome maps of these uptake ratios are provided in Figure 2.7D (86-028NP) and 2.7G (PittGG;) they display the consistency of the peak heights across each genome.  As expected, the long-fragment DNA samples (Figure 2.6B and Figure 2.7B, E and H) had much less variation in uptake ratio than the short-fragment samples; 90% of positions had uptake ratios within two-fold of the mean, and there were few high peaks or low valleys. The few extended segments with low or no uptake coincided with large gaps between USS10s. The largest gap was in the 86-028NP segment between 95 and 145kb —the site of a genomic island that is absent from the PittGG and Rd strains, has few USS and has high similarity to an H. influenzae plasmid (Harrison et al., 2005). However, uptake ratios did exhibit substantial short-range variation not predicted by the model, which is considered further below. 34  Figure 2.6. Experimentally determined uptake ratios for a 50 kb segment. The X-axis is the same 50 kb segment of the 86-028NP genome as Figure 2.3G and H. Grey points indicate positions with input coverage lower than 20 reads. Gaps indicate unmappable segments. A. Uptake ratios of short-fragment DNA. Inset: Same data with a logarithmic-scale Y-axis. B. Uptake ratios of long-fragment DNA. C. Locations and scores of USS10s.  0-2-9-12-USS10scoreA.B.C.10000 20000 30000 40000 5000004-6-2-8-010-210-410 50kb4-Short DNA uptake ratioLong DNA uptake ratio86-028NP genome position 35  Uptake Ratio (short DNA)Uptake Ratio (long DNA)USS10scoreI.02460246912PittGG genome position (bp)500000 1000000 15000000G.H.Uptake Ratio (short DNA)Uptake Ratio (long DNA)USS10score0246024912PittGG genome position (bp)10000 20000 30000 40000 500000A.B.C.Uptake Ratio (short DNA)Uptake Ratio (long DNA)USS10score024602491286-028NP genome position (bp)500000 1000000 15000000D.E.36 Figure 2.7. Experimentally determined uptake ratio maps and USS10 maps. Grey points indicate positions with input coverage lower than 20 reads.  Gaps indicate unmappable positions.  A-C:  Maps of a 50 kb segment of the PittGG genome.  A. Uptake ratios of short-fragment PittGG DNA.  B. Uptake ratios of long-fragment PittGG DNA.  C. PittGG USS10 positions and scores.  D-I: Whole-genome maps.  D. Uptake ratios of short-fragment 86-028NP DNA.  E. Uptake ratios of long-fragment 86-028NP DNA.  F.  86-028NP USS10 positions and scores.  G. Uptake ratios of short-fragment PittGG DNA.  H. Uptake ratios of long-fragment PittGG DNA.  I.  PittGG USS10 positions and scores.  37   Figure 2.8.  Shapes of typical 86-028NP uptake peaks. Blue dots: uptake ratios after smoothing with a 31 bp window. A.-H.: Orange dots: uptake ratios without smoothing (note that Y axis is offset by 0.5 units). Red triangles and numbers: locations and scores of USS.  A.-F. 86-028NP 6-4-2-0-Smoothed uptake ratio -4-2-0 Raw uptake ratio6-4-2-0-Smoothed uptake ratio6-4-2-0-Smoothed uptake ratio6-4-2-0-Smoothed uptake ratio100-0-Read coverage0 2 4 6 70900 71150 71400 70900 71150 71400B.12.20 2 4 6 31000 31250 31500 31 00 31 0 31500A.0 2 4 6 1045050 1045300 1045550 9.81045050 1045300 1045550C.1837250 1837500 18377500 2 4 6 1837250 1837500 1837750 D.10.4548300 548550 5488000 2 4 6 548300 548550 548800 F.11.511.70 2 4 6 457100 457350 457600 12.0 12.3E.457100 457400 457600-4-2-0 Raw uptake ratio-4-2-0 Raw uptake ratio-4-2-0 Raw uptake ratio1226850 1227200 122745018250 18500 187500 2 4 6 1226950 1227200 1227450 H.0 2 4 6 18250 18500 18750 G.0 100 1226950 1227200 1227450 J.0 100 18250 18500 18750 I.86-028NP genome position (bp)11.538 short-fragment DNA: A. and B.: peaks at strong USS. C. and D.: peaks at weak USS. E. and F.: peaks at pairs of USS separated by A. 69 bp and B. 230 bp. G. and H. Uptake ratio spikes not at USS in 86-028NP long-fragment DNA. I. and J.  Purple dots: sequencing coverage of input 86-028NP long-fragment DNA. Sensitivity is limited by low sequencing coverage: At some genome positions our ability to detect USS-dependent uptake biases and possible USS-independent biases in uptake coverage was limited by low sequencing coverage. Although some of this low coverage arose from the contamination-correction step described above, segments of low coverage were also seen in the raw data, likely arising from biases in the library preparation and sequencing steps. Figure 2.9 compares the raw coverage of short and long input samples for a 50 kb segment of the 86-028NP genome, illustrating the strong variation in sequencing coverage that was both broadly reproducible and sequence dependent. Low coverage had similar effects in all samples, precluding calculation of uptake ratios where input coverage was zero and generating high levels of stochastic variation where coverage was low (segments with no coverage and positions with coverage below 20 reads are indicated in uptake ratio maps by gaps and grey dots respectively).  Figure 2.9.  Variation in read coverage. Read coverage of the 86-028NP long-fragment (green) and short-fragment (purple) input samples over a 50 kb genome segment. We used the low uptake ratios of positions at least 1000 bp from the nearest USS9.5 to assess the effectiveness of the contamination correction described above, since these positions are where 86-028NP genome position (bp)Read coverage depth39 contamination would have had its largest effects. This peak-separation distance was chosen because a peak-shape analysis of uptake around 158 USS10s that were separated by at least 1200 bp from other USS10s and had uptake ratios of at least 3.0 found that mean uptake ratio had fallen to baseline at positions 600 bp from the USS (Figure 2.10). Figure 2.11 compares uptake ratios between the valley positions that received the contamination correction described above (those with a Rd homolog) and the positions where no correction was needed (those with no Rd homolog). The two distributions had nearly identical medians (0.0022 and 0.0020 respectively), indicating that the correction was sufficient but not excessive. These low values also confirm that the DNase I treatment and washing steps removed at least 99.99% of the donor DNA that had not been taken up.  Figure 2.10.  Shape analysis of isolated USS10 peaks. Short-fragment uptake ratio data for positions around 158 86-028NP USS10s that were separated by at least 1200 bp from other USS10s and had uptake ratios of at least 3.0. Distance from USS center (bp)200-200 400-400 600-600 0|         |         |         |         |         |         |         |         |         |         |         |         |    |         | Count1 1000|         |         |         |Uptake ratio|                    |                    |                    |                    |                    |                   | 01.02.03.04.05.06.040     Figure 2.11. Depths of uptake ratio valleys with and without contamination correction. Uptake ratios of positions in the 86-028NP short-fragment dataset that were at least 1kb from the nearest USS9.5 and whose uptake coverages were either corrected for contamination with homologous Rd sequences (left, 272,547 positions) or not corrected for contamination because they had no Rd homology (right, 89,418 positions). Orange and red dashed lines indicate valley uptake predicted by the original and revised models respectively.  Uptake ratios show no periodicity across the genome: Bacterial genomes show periodicity for several features related to DNA curvature and codon usage biases (Mrazek, 2010), so we examined the distribution of uptake ratios across each genome by Fourier analysis, using the R package TSA. The log-log plots in Figure 2.12 show that this found no strong influence of any specific repeat period on either the variation in input-sample coverage (panels A-D) or the variation in uptake ratios (panels E-H). Instead, to explain the observed variation, the analysis needed to invoke small contributions from almost every possible repeat period.  Uptake ratio10-1 -With Rd homologyWith no Rd homology10-3 -10-5 -41  Figure 2.12. Tests of periodicity. Fourier-transform analyses were performed using R-package RCA.  The X-axes are log10 of repeat period in bp; Y-axes are log10 of the relative periodicity at each repeat period.  A-D.  Tests using coverage in input samples.  E-H. Tests using uptake ratios. Samples: A & E: 86-028NP short fragments; B & F: 86-028NP long-fragments; C & G:Pitt GG short fragments; D & H: PittGG long fragments.   Relative power1,000,000 10,000 100Relative powerRelative power1,000,000 10,000 100Period of repeat (bp) Period of repeat (bp)Relative power1,000,000 10,000 1001,000,000 10,000 100Period of repeat (bp) Period of repeat (bp)F.10-510-110310-510-110310-510-110-510-1D.C.B.A.E.G. H.F.42 Uptake of short-fragment 86-028NP DNA To investigate how USSs contribute to the DNA uptake process, our strategy was to first analyze discrepancies between model predictions and observed uptake ratio peaks in the 86-028NP short-fragment dataset, since these would reveal ways in which the simple assumptions underlying the model mischaracterized the actual steps of DNA uptake. Model changes that improved the predictions were considered to better reflect the true constraints on uptake of short DNA fragments. The refined model’s predictions were then compared to the real uptake ratios for 86-028NP long-fragment DNA, allowing additional refinement, and finally to the short- and long-fragment uptake ratios for PittGG DNA.  Figure 2.13 compares predicted and measured uptake (orange and blue lines respectively) of short-fragment 86-028NP DNA for the first 50kb of the genome. The close correspondence between the model’s predictions and observed peak locations and shapes confirmed that almost all of the variation in uptake was due to USSs. (The Pearson correlation coefficient over all positions was 0.92). However, the more detailed analyses below identified components of the model that could be improved. First, the predicted low-uptake valleys were too high. This is not easily seen in Figure 2.13, but the log-scale inset in Figure 2.6 and the analysis in Figure 2.11 show that most valley-bottom positions had uptake ratios between 0.01 and 0.001, well below the model’s predicted baseline uptake of 0.052 (dashed orange line in Figure 2.11). Lowering the baseline of our predicted model increases the uptake peak’s heights but this effect was very small. For instance, decreasing the baseline of our initial model 100-fold (from 0.05 to 0.0005), increased peaks height by only 3.8%.    43  Figure 2.13. Predicted and observed DNA uptake analysis for short fragments of 86-028NP DNA. A. Map of uptake ratios and initial model predictions. The blue points show the same uptake ratio map as in Figure 2.6A. The orange points show the same predicted uptake as in Figure 2.3G. B. Map of differences between observed minus expected uptake ratios. C. Locations and scores of USS9.5s.  Second, the score cutoff of 10.0 bits was too high. Analysis of score-subsets within the uptake-valley dataset showed that 9.5 bits was a better cutoff.  For the 78 positions with scores between 9.5 and 10.0 bits, the correlation between score and uptake ratio was 0.27.  Although a similar correlation (0.27) was seen for the 136 positions with scores between 9.0 and 9.5 bits, it was driven mainly by very small effects, and only 4 positions had uptake ratios higher than 0.1. For the 461 positions with USS scores between 8.5 and 9.5 bits the correlation was only 0.09. Third, most of the predicted peaks were too high. Figure 2.14 complements Figure 2.11’s analysis of uptake valleys with an analysis of similarly isolated uptake peaks. The blue dots show Uptake of short-fragment DNA9-12-USS10scoreC.10000 20000 30000 40000 50000086-028NP genome position 4 -6 -2 -0 -A.8 -0 -4 --2 -2 -B.DifferenceObserved from expected uptake44 the uptake ratio at every USS9.5 position in the genome that is at least 1000 bp from the nearest USS10 (n=209), and the small orange dots show predicted uptake at the same positions. The lack of scatter in the orange points confirms that the separation distance was sufficient to avoid predicted effects of nearby USS; the least-squares difference between predicted and observed uptake ratios at these 209 USS positions was 0.521.  Figure 2.14. Short-fragment uptake ratios and predicted DNA uptake at isolated 86-028NP USS as a function of USS score. Predicted or measured DNA uptake at 209 USS9.5 positions separated by at least 1000 bp from the nearest USS10s. The blue dots show the measured uptake ratios; grey bars show the ranges of the three replicates at each position. The blue line shows a sigmoidal function fit to these points. The small orange dots and line show uptake predicted by the original model at the same positions, and the small purple and red dots show uptake predicted by the intermediate and revised model versions discussed in the text.  In the original-version of the uptake model, the sigmoidal parameters of the p_uptake function (baseline and location and slope of inflection point) were set using the frequencies of USS scores in the 86-028NP genome; as expected, the same parameters were obtained when a sigmoidal function was fit to the uptake points predicted by this model (orange line in Figure 2.14). To USS score (bits)Uptake ratio or predicted uptake at USS012345610 11 12 13945 improve the predictions we replaced these p_uptake parameters with those of a sigmoidal function fitted to the real uptake data (blue line in Figure 2.14), changed the p_uptake baseline to 0.005, lowered the USS score cutoff from 10.0 bits to 9.5 bits, and ran the model again. The purple points in Figure 2.14 show that, although peak-height predictions for low-scoring USS were somewhat improved by the changed p_uptake function, those for USS with scores > 11.5 bits became slightly worse; the least-squares difference for the 209 positions was modestly reduced from 0.521 to 0.420. We initially suspected that the over-prediction of peak heights might be due to overestimating the proportion of very short fragments in the input DNA, but eliminating the contributions of fragments shorter than 80 bp had very little effect on predicted peak heights. (This is likely because these fragments contribute little DNA and rarely contain USS.) We then considered an alternative explanation, that USS might be ineffective when they were very close to fragment ends. The red dots in Figure 2.14 shows that requiring USS to be at least 50 bp from fragment ends lowered predicted peak heights to the mean observed height while maintaining a low baseline uptake. Distances of 30 bp and 70 bp were also tested, but gave peaks that were, respectively, too high and not high enough. Incorporating this modification into the model barely changed dramatically reduced the least-squares difference of its predictions with the observed uptake ratios (0.93, up from 0.92), but dramatically reduced the least-squares difference from 0.420 to 0.052. This revised model was used by of the analyses described below.  The above analyses do not explain why many uptake ratio peaks were substantially higher or lower than predicted by their USS score. The scatter is unlikely to be due to noise alone, since more extreme smoothing of the uptake ratio data with windows as large as 150 bp improved the correlation by only 0.001.  Below we consider three other factors that might influence DNA uptake: weak non-USS uptake biases, effects of interactions between bases at different positions in the USS, and effects of DNA shape. Weak uptake biases could arise from either low-scoring USS or non-USS sequence factors. Since weak biases would only be detectable in genome segments that lack strong USSs, we searched for them using a far-from-USS10 dataset containing only DNA segments whose ends were at least 0.6kb from the closest USS10 and that had input coverage of at least 20 reads. This dataset 46 contained 513 segments where weak uptake effects could in principle be detected (22% of the genome); their mean uptake ratio over the 428678 positions was 0.0102. In these segments, the only positions with distinct uptake ratio peaks >0.2 were 9 weak USSs with scores between 9.5 and 9.9 bits (an example is shown in panel C of Figure 2.8). Since this analysis did not find any non-USS positions giving uptake higher than 0.2, it suggests that other sequence factors do not detectably promote uptake in the absence of a USS. The degenerate-USS analysis of Mell et al. (Mell et al., 2012) found that pairwise interactions between the AT-tract bases and core bases of a USS made substantial contributions to uptake of a 200 bp synthetic fragment. To evaluate the effects of these interactions in predicting uptake of genomic DNA, the USS scores used by our revised model were raised or lowered in proportion to the interaction effects reported in Figure 2.14 of Mell et al. Figure 2.15 shows that, for the same 209 isolated USS9.5s used above (Figure 2.14), this adjustment had little effect on high-scoring USS but further reduced the scores of low-scoring USS. The data points are coloured by their uptake ratios, showing that weak USS giving unusually high or low uptake were equally likely to have their scores reduced. However, these scoring changes had no effect on the short-fragment uptake predictions of the revised model (both Pearson correlations 0.93), probably because (i) the scores of strong USS were not significantly changed and (ii) the scores of weak USS already gave near-baseline uptake.  47  Figure 2.15. Effects of within-USS interactions on USS scores. USS scores calculated with and without interactions effects for 209 isolated 86-028Np USS9.5s (see Methods for details). Red line shows expected scores if the interactions had no effect. Point color indicates the uptake ratios at the USS: yellow, <0.01; red, 0.01-0.1; purple, 0.1-0.5; blue 0.5-1.5; green, >1.5.   DNA shape effects: Most of the interactions in Figure 2.15 were between bases separated by 10 bp or more, so we complemented that analysis with analysis of DNA shape, which reflects both pairwise and more complex interactions over a 5 bp range. The major shape features that can be predicted from DNA sequence are the minor groove width, the propeller twist between two paired bases, the helix twist between one base pair and the next, and the roll of one base pair relative to the next. In Figure 2.16, the wide grey line reproduced in each panel shows these feature predictions for the consensus USS. The USS inner core (orange shading) has a relatively wide minor groove and high propeller twist, which would facilitate sequence recognition by proteins (Rohs et al., 2009). To the left of this and in both AT-tracts (yellow shading) the minor USS score without interaction effects (bits)10 11 12 139USS score with interaction effects (bits)101112139848 groove is narrow with low propeller twist and negative roll, predicting that these segments are both rigid and slightly bent.  To see if shape features affect uptake in ways that were not captured by scores alone, we compared the shape features of subsets of the 209 isolated USS with similar scores but different uptake ratios. Panels A-D of Figure 2.16 compare the shape features of very weak USS (USS9.5-10) whose uptake ratios were either low (<0.6, blue lines) or high (>2.0, orange lines). Similarly, panels E-H and I-L show the same comparisons for USSs with low and moderate scores (USS10-10.5 and USS10.5-11 respectively). USSs with scores higher than 11 bits were not analyzed since they did not exhibit enough uptake variation to reveal correlations between uptake and DNA shape.  In all three score subsets the inner-core shape features were very similar for low-uptake and high-uptake subsets (blue and orange lines), probably because this sequence perfectly matches the consensus in 206 of the 209 USS9.5. (The other three had otherwise-perfect core sequences and good AT tracts but had only baseline uptake ratios.) However, the AT-tract shapes had marked differences, with low-uptake USSs having no distinctive shape features and high-uptake USSs resembling the USS consensus shape. This suggests that the predicted rigidity and slight bend caused by interactions within the consensus AT tracts facilitate DNA uptake.  49  Figure 2.16. Predicted shape features of USS. The thick grey line reproduced in each box shows shape analysis of the consensus USS sequence. Blue and orange lines in each box show shape analysis of genomic USS separated by at least 500 bp, grouped by score and coloured by uptake ratio. The orange and yellow bars below each box indicate components of the USS (see Figure 1): light orange: outer core; dark orange: inner core; yellow: AT tracts. Panels A-D USS9.5-10: Blue: uptake ratios <0.2 (n=68, mean USS score =9.7). Orange: uptake ratios >0.2 (n = 10, mean USS score =9.8). Panels E-H: USS10.0-10.5: Blue: uptake ratios <0.6 (n=47, mean USS score=10.22). Orange: uptake ratios >2.0 (n=10, mean USS score=10.26). Panels I-L: USS10.5-11.0: Blue: uptake ratios <0.6 (n=14, mean USS score=10.64). Orange: uptake ratios >2.0 (n=59, mean USS score=10.79). A, E and I. Minor groove width, in Å. B, F and J. Propeller twist, in degrees. C, G and K. Helix twist, in degrees. D, G and L. Base pair roll, in degrees. See also Figure S5. Grey bars for each point show the standard error.  Dots indicate significant differences by Kolmogorov-Smirnov between high-uptake and low-uptake positions with (red dots) and without (black dots) Bonferroni correction. Minor groove width (Å)Propeller twist (°)Roll (°)USS position (bp)E.F.G.H.I.J.K.L.Helix twist (°)A.B.C.D.ı11ı3ı19ı27ı11ı3ı19ı27ı11ı3ı19ı272 -4 -6 --15 --5 --32 -38 -35 -0 -4 -8 -USS10.0-10.5 USS10.5-11.0USS9.5-10.050 Uptake of fragments with more than one USSs: Many genomic USSs are sufficiently close that they will co-occur even on short DNA fragments; 23% of 86-028NP USS10s are within 100 bp of another USS10, and 17% are within 30 bp (Figure 2.17A). Fragments with multiple USS might be expected to have relatively high uptake, since they provide more targets to which the uptake machinery receptor could bind, but only one of the two previous studies in Neisseria found this effect (Ambur et al., 2007; Goodman and Scocca, 1991).  Visual examination of uptake ratio maps around the 230 pairs of 86-028NP USS10s within 100 bp found only single peaks; Figure 2.17B shows that the midpoints of these USS pairs (red and green points) have very similar uptake ratios to those of the 209 isolated USSs (pale blue points). Figure 2.8 panels E and F show examples of peaks at USS pairs separated by 69 bp and 230 bp. A special subclass of USS pairs consists of oppositely oriented pairs that are so close that they overlap; these can form RNA hairpins and are usually located at the ends of genes where they act as transcriptional terminators (Kingsford et al., 2007; Smith et al., 1995, 1999). Figure 2.17A shows that the 86-028NP genome has 109 USS10 pairs whose centers are within 14 bp: 69 in the -/+ orientation, all 0-3 bp apart, and 40 in the +/- orientation, all 10-14 bp apart. The green points in Figure S9B shows that midpoint uptake ratios of these pairs were very similar to those of other pairs or at isolated USS10s with similar scores.  51  Figure 2.17.  Analysis of DNA uptake effects of USS10 pairs in the 86-028NP genome. Frequencies of spacings between close USS10 pairs. B. Uptake ratios for isolated USS10 (blue points, data from Figure 2.14) and for centers of pairs of USS10 whose centers are 14-100 bp apart (red points) or 0-13 bp apart (green points). Overall, these results indicate that the presence of two USS10s within 100 bp does not detectably increase the probability of the receptor binding to a USS, a result consistent with that of Ambur et al. for pairs of closely spaced DUS in Neisseria meningitidis (Ambur et al., 2007). Because individual fragments were not tracked, we cannot make any conclusions about uptake of fragments with more widely separated USS. Uptake of long-fragment 86-028NP DNA Figure 2.18A compares the revised model’s predictions for long-fragment 86-028NP DNA with the uptake ratios observed over the same 50kb genome segment as in Figure 2.6B. In contrast to predictions for short-fragment uptake (correlation of 0.93), it seriously underpredicted the USS score (isolated USS) or mean USS score (pairs of USS)A.B.[-0.1, 0.9](0.9, 1.9](1.9, 2.9](2.9, 3.9](3.9, 4.9](4.9, 5.9](5.9, 6.9](6.9, 7.9](7.9, 8.9](8.9, 9.9](9.9, 10.9](10.9, 11.9](11.9, 12.9](12.9, 13.9](13.9, 14.9](14.9, 15.9](15.9, 16.9](16.9, 17.9](17.9, 18.9](18.9, 19.9](19.9, 20.9](20.9, 21.9](21.9, 22.9](22.9, 23.9](23.9, 24.9](24.9, 25.9](25.9, 26.9](26.9, 27.9](27.9, 28.9](28.9, 29.9](29.9, 30.9](30.9, 31.9](31.9, 32.9](32.9, 33.9](33.9, 34.9](34.9, 35.9](35.9, 36.9](36.9, 37.9](37.9, 38.9](38.9, 39.9](39.9, 40.9](40.9, 41.9](41.9, 42.9](42.9, 43.9](43.9, 44.9](44.9, 45.9](45.9, 46.9](46.9, 47.9](47.9, 48.9](48.9, 49.9](49.9, 50.9](50.9, 51.9](51.9, 52.9](52.9, 53.9](53.9, 54.9](54.9, 55.9](55.9, 56.9](56.9, 57.9](57.9, 58.9](58.9, 59.9]010203040506070All -/+All +/-Uptake ratio or mean uptake ratio9.5 10.0 10.5 11.0 11.5 12.0 12.56420Number of USS10pairs 60402000 10 20 30 40 50 60Separation of USS centers (bp)52 variation in long-fragment uptake ratios (correlation of 0.60). Although 51% of 86-028NP positions had long-fragment uptake ratios lower than 0.8 or higher than 1.2, only 13% had predicted uptake outside these limits.   Figure 2.18. Predicted and observed uptake of long 86-028NP DNA fragments. A. and C. Uptake maps for 86-028NP long-fragment DNA. Orange points: USS-dependent uptake predicted by the revised model. Blue points: mean uptake ratios from 3 replicate experiments (grey indicates input coverage <20 reads, gaps indicate unmappable positions). B. and D. USS9.5 positions and scores. A. and B. The same 50 kb genome segment shown in previous figures. C. and D. Whole genome. Because long-fragment uptake ratios lacked the dramatic peaks and valleys seen for short-fragments, stochastic noise arising at the regions of low sequencing coverage described earlier was expected to play a larger role. To estimate the magnitude of this effect, we compared the effects of adding different amounts of artificially generated noise to simulated (noise-free) uptake data (Figure 2.19A). Figure 2.19B shows that, as expected, the correlation between noisy A.B.420Uptake ratio9.012.6USS9.5score|10000|20000|30000|40000|50000|086-028NP genome position C.D.9.012.6USS9.5scoreUptake ratio240|500000|1000000|1500000|086-028NP genome position 53 and noise-free data worsened as the arbitrary level of noise increased for both short-fragment (blue) and long-fragment (red) simulations, and that increasing noise had a much stronger effect on the long-fragment simulations. For short DNA fragments, simulations with noise levels of 2 and 2.5 gave correlations of 0.94 and 0.92, very close to the 0.93 correlation between the revised model and the real data. (dashed blue line in Fig. 2.19B) For long DNA fragments with the same noise levels, more noise was needed, with a 3.0 multiplier giving a correlation of 0.56, slightly lower than the observed correlation of 0.60 (dashed red line in Fig. 2.19B). This confirms that the disparities between measured uptake ratios and USS-based predictions were at least partially due to noise in the sequencing data. However, these correlations are still 21% and 10% higher than the best correlation obtained between the revised model’s predictions and real data.   Figure 2.19. Simulated noise analysis. A. Effects of added red noise on simulated noise-free coverage. Black points: no added noise; Green, yellow and red points: simulated noise added Simulated noise multiplierPearson correlation coefficient0 1.0 2.0 3.01.00-0.75-0.50-B.Simulated coverageGenome position (bp)200-100-0-|15,000|17,000|16,000| |A.54 with multipliers of 1.0, 2.0 and 3.0 respectively. B. Correlation coefficient from simulated uptake ratios with and without different levels of noise. X-axis represents the multiplicative factor applied to the coverage-dependent amount of noise added. Simulated results for 86-028NP-short and 86-028NP-long DNA fragments are shown in blue and red, respectively. Dashed lines at 0.93 (blue) and 0.60 (red) indicate the real-data correlation of predicted uptake with observed uptake ratios for 86-028NP-short (top) and long (bottom) fragment size distributions, respectively.   One notable feature of the long-fragment uptake ratio maps in Figure 2.6B and 2.18A is the presence of occasional spikes of unusually high uptake, e.g. at position 18,540 (examples are shown in panels G and H of Figure 2.8). These spikes were much narrower than expected for true uptake biases acting on long DNA fragments, so they were likely due to stochastic differences between input and uptake coverage in regions of low coverage, not to true differences in uptake. Consistent with this explanation, 70.4% of positions with uptake ratios greater than 2.0 had coverage less than 100 reads, compared to only 8% of positions with more typical uptake ratios between 0.5 and 2. Table S3 shows analysis of the distribution of uptake extremes at positions with different levels of sequencing coverage, and panels I and J of Figure 2.8 show the low coverage at the example spikes. However, this explanation did not apply to most positions with very low uptake, which were in broad segments with few or weak USS, not in narrow spikes at low coverage positions, and thus likely reflect genuinely low uptake.  Prediction of uptake of PittGG DNA Since the uptake model was refined using uptake data for DNA of strain 86-028NP, the revised model’s predictions were assessed using uptake data for DNA of a different strain, PittGG, which differs from 86-028NP DNA by SNPs and indels affecting about 11% of its genome. Figure 2.20 compares the uptake predictions with the observed PittGG uptake ratios. For short-fragment data the peak heights and valley depths were similar to those for 86-028NP, as was the Pearson correlation between predicted and observed uptake (0.92). Although the model predicted similar long-fragment variation for 86-028NP and PittGG, the PittGG experimental data showed more extreme variation, and the correlation was only 0.41, substantially worse than the 0.60 obtained for 86-028NP. As with the 86-028NP data, much of this discrepancy may be due to noise in 55 regions of low sequencing coverage. However, this may not fully explain PittGG’s lower correlation, because input samples of the two strains had similar frequencies of low-coverage positions (Table S3).  56  A.C.0 2000010000 30000 40000 50000USS score12.690 1000000500000 1500000USS score12.69E.G.0 2000010000 30000 40000 50000USS score12.690 1000000500000 1500000USS score12.69Predicted or observed uptake ratioPredicted or observed uptake ratioPredicted or observed uptake ratioPredicted or observed uptake ratioB.D.F.H.24624624624657 Figure 2.20 Predicted and observed uptake ratios for PittGG short and long DNA fragments.  A and E: Uptake maps for the first 50kb of the PittGG genome; C. and G.: Uptake maps for the full PittGG genome. B., D., F. and I.: Vertical tick marks indicate locations and scores of USS10s.  Orange lines: USS-dependent uptake predicted by the revised model.  Blue lines: Mean uptake ratios from 3 replicate experiments (grey points indicate positions with <20 reads input coverage, gaps indicate unmappable positions). Note that some grey dots are beyond the tops of some panels. Predicted competition with other DNAs in the human respiratory tract H. influenzae’s natural environment is the human respiratory tract, where its DNA must compete for uptake with DNAs from other bacteria, and with host-derived DNA whose concentration in respiratory mucus of healthy individuals can exceed 300 µg/ml (Lethem et al., 1990; Shak et al., 1990). The revised model was used to investigate this competition. Table 2.2 lists the frequencies of USS10 and USS11 in the genomes of various species. Pasteurellaceae species that share H. influenzae’s Hin-type USS (e.g. H. parainfluenzae and Aggregatibacter actinomycetemcomitans) typically have about 1000 USS10s per Mb (Redfield 2006); most of these are strong USS with scores ≥ 11. Pasteurellaceae with the Apl-type USS (e.g. H. ducreyi and Mannheimia haemolytica) have about tenfold fewer Hin-type USS10s, and fewer than 20% of these are strong. Other common respiratory tract bacteria (e.g. N. meningitidis, Streptococcus pneumonia and Pseudomonas aeruginosa) have USS only at the frequencies expected for their base compositions. The human genome is exceptional in having about 5-fold fewer USS10 per Mb than expected from simulated sequences with the 41% GC content of human DNA (observed frequency 4.6/Mb, simulated frequency 30/Mb, both with mean scores of only 10.3 bits). Since the USS inner-core motif includes a CpG, the underrepresentation of USS10 in human DNA relative to simulated sequences is probably a consequence of the 4-5 fold depletion of CpGs in the human genome caused by deamination of methylated cytosines (Babenko et al., 2017; Bird, 1986).  Table	2.2.		Frequencies	of	Hin-type	USS	in	genomes	of	other	species	Species Genome size USS10/Mb USS11/Mb H. influenzae 86-028NP 1.914 1014 754 58 H. parainfluenzae T3T1 2.087 867 683 Aggregatibacter actino-mycetemcomitans VT1169 2.129 970 737 H. ducreyi 35000HP 1.699 102 16 Mannheimia haemolytica M42584 2.732 96 18 N. meningitidis 2.272 52 4 Streptococcus pneumoniae 2.039 16 1 Pseudomonas aeruginosa 6.264 8 0 Homo sapiens1 3 x 1.9 Mb 6.6 0.1 Random-sequence DNA2, 41% G+C 3 x 1.9 Mb 29.5 0.3 1. Means for three 1.9 Mb segments of human chromosomes 1, 3 and 12.   2. Means for three 1.9  Mb ‘genomes’ with the same base composition as human DNA, described in Methods.  To estimate how strongly these DNAs would compete with H. influenzae DNA for uptake by H. influenzae cells, we concatenated the 86-028NP genome with each of two Pasteurellacean genomes and with three 1.9 Mb segments of the human genome, and used the revised uptake model to predict relative uptake. The respiratory pathogen H. parainfluenzae was used as the 59 Hin-type USS representative. The genital pathogen H. ducreyi was used as a representative species with Apl-type USSs, since none occur in the human respiratory tract, and segments of human DNA also served to represent non-Pasteurellacean bacteria. To approximate the lengths of DNA fragments in the respiratory tract (Lethem et al., 1990; Shak et al., 1990), the model was run using fixed fragment lengths of 1kb and 10kb, and because human DNA will contain many fragments lacking USSs, the predictions were made with and without the model’s p_uptake baseline binding probability of 0.005. For each genome in the concatenated sequence, the predicted uptake at every position was then summed to get a total genome uptake value for each fragment length and baseline assumption.  Table 2.3 shows the relative amounts of H. influenzae DNA predicted to be taken up under the various competition conditions. As expected from published uptake-competition experiments (Albritton et al., 1984; Redfield et al., 2006), H. influenzae and H. parainfluenzae DNAs were taken up with equal efficiency in all simulated conditions, and H. ducreyi DNA was much less competitive, contributing only 5% of the 1 kb fragments and only 12% of the 10 kb fragments. With human DNA as the 1:1 competitor, more than 99% of the DNA taken up was predicted to be from H. influenzae. For all competing DNAs, baseline uptake of fragments containing no USS made only a tiny contribution. Table	2.3.	Predicted	relative	uptake1	of	H.	influenzae	DNA	in	simulated	competition	with	DNAs	of	other	species2.		Assumptions:  1 kb fragments, p_uptake=0 1 kb fragments, p_uptake =0.005 10 kb fragments, p_uptake=0 10 kb fragments, p_uptake=0.005 86-028NP DNA in competition with:     H. parainfluenzae DNA (Hin-type USS)  0.500 0.500 0.475 0.475 H. ducreyi DNA (Apl-type USS)  0.952 0.951 0.884 0.883 60 Homo sapiens DNA3 (no USS enrichment) 0.998 0.997 0.991 0.990 1. H. influenzae DNA as a fraction of the total DNA predicted to be taken up. 2. Competitions were genome:genome 3. Means for the three genome 1.9 Mb segments described in Methods. Discussion DNA uptake by competent H. influenzae Rd cells was measured at every position in the genomes of two divergent H. influenzae strains, using short-fragment and long-fragment DNA preparations. Differences between observed uptake and that predicted by a computational model of USS-dependent uptake revealed the strength of the uptake machinery’s bias towards USS and the absence of other sequence biases. These findings increased our understanding of DNA uptake bias and its potential effects on the distribution of recombination.  Implications for DNA uptake The USS motif: The measured discrimination’ for USS was very strong; with short DNA fragments, valleys at USS-free segments had ~1000-fold lower uptake ratios than peaks at high-scoring USS. This non-zero baseline is unlikely to be due to residual contamination by recipient DNA, since valley depths were similar for segments with and without Rd homology (Figure 2.11). One surprising finding of Mell et al.’s (2012) work was the difference between the inner-core uptake motif identified by their degenerate-USS experiments (Figure 2.1A) and the extended motif of USS sequences in the H. influenzae genome (Figure 2.1B). Our analysis confirmed that uptake absolutely requires a perfect match to the inner core, but found that this was not sufficient to raise uptake above baseline, even if the rest of the core was perfectly matched.  Effect of USS shape: The predicted shape differences between similarly-scoring USSs that gave strong or weak uptake (Figure 2.16) suggest a preference for USS that are rigidly bent at AT-tracts and the outer core (Harteis and Schneider, 2014; Rohs et al., 2009). Similar preferences have been described for several DNA binding proteins and have been associated with specific binding by arginine or lysine residues to narrow minor grooves (Rohs et al., 2009; Stella et al., 61 2010). These features have been integrated successfully in some transcription factor binding models (Li et al., 2017), but using them to improve uptake prediction will require more comprehensive investigation into the effects of DNA shape on uptake. The stiffness also suggests that the initial passage of DNA through the secretin pore may be facilitated by transient strand melting at or beside the USS rather than by bending (Danner et al., 1982). Other USS effects on DNA uptake: Comparison of predicted and measured heights of uptake peaks suggested that USSs were ineffective when located very close to fragment ends. This would be consistent with experimental evidence that uptake initiates internally rather than at fragment ends, but would need to be experimentally investigated.  The presence of two USSs within 100 bp did not detectably increase uptake, a finding consistent with Ambur et al.’s (Ambur et al., 2007) study of very close uptake sequences in Neisseria. Lack of USS-independent effects: The very low valleys between short-fragment uptake peaks allowed us to examine more than 400,000 positions for USS-independent increases in uptake. None were found; the only positions with distinct uptake ratio peaks >0.2 were 9 weak USSs. Implications for genetic exchange Preferential uptake of USS can create variation in recombination at all levels: across a single genome, between strains of one species, and between both closely related and unrelated species. Pifer & Smith showed that transformation frequency in H. influenzae decreased exponentially when fragments were smaller than 3.5 kb. This decrease was attributed to exonuclease degradation of fragments from their 3’ ends, since similar numbers of short and long fragments were taken up, only 5’-end label was incorporated into the chromosome, short fragments transformed more efficiently when the selected marker was far from one end. Mell (2014) used genome sequencing of inter-strain recombinants to examine the distribution of recombination tract lengths; despite the presence of 2-3% SNPs, the mean tract length was 6.9 kb and the longest was 43 kb. The efficient uptake and recombination of long fragments allows fragments containing non-homologous segments to still recombine well, provided both fragment ends are homologous. Fragments with only one homologous end recombine much less efficiently (‘homology-facilitated recombination’ (de Vries and Wackernagel, 2002)), and integration of fragments with no homology is very rare. 62 Across the genome:  Across the genome of a H. influenzae strain, the genetic consequences of USS-dependent DNA uptake depend on USS locations and on the lengths of the available DNA fragments. If only short fragments are available, the limitation to positions close to a USS may be obscured by limitation caused by degradation of incoming DNA in the cytoplasm. If most fragments are long, recombination will be both more frequent and more evenly distributed across the genome, because long fragments are more likely to both contain USS and recombine. The result will be that almost all recombination is caused by fragments long enough to usually contain at least one strong USS. This situation is caused by the high abundance and relatively even distribution of genomic USS, itself an expected consequence of the functionally neutral accumulation of USS under a molecular drive caused by USS-biased DNA uptake (Danner et al., 1982; Maughan et al., 2010). Recombination between H. influenzae strains: On average about 85% of the genomes of H. influenzae strains are homologous, with sequence divergence low enough to have only a modest effect on recombination frequencies (Mell et al., 2011). On average, genome segments that are absent from other strains have lower density of USS (0.58/kb vs 1.08/kb for sequences present in 86-028 but absent from Rd. Only some of these will be segments newly acquired from species without USS. If a non-homologous segment introduced into one strain by conjugation or transduction is beneficial, the USSs in adjacent DNA will help it efficiently spread to other strains by transformation. Recombination between Pasteurellaceae species: Uptake of DNA from related species can also influence recombination, either directly if the DNA is sufficiently similar to recombine with the H. influenzae genome, or indirectly if it successfully competes with H. influenzae DNA for uptake or, once inside the cell, for access to nucleases or recombination machinery.  For H. influenzae the most important competition will be with other Pasteurellaceae that share both the respiratory tract niche and the Hin-type USS, but similar effects are expected for Pasteurellaceae in other host species.    63 Competition with DNAs from human cells and other respiratory bacteria: In the respiratory tract, the most important source of competing DNA is human cells. However, our analysis suggests that H. influenzae’s uptake specificity allows its DNA to outcompete human and other foreign DNAs even if these are in 100-fold excess, a combined effect of the low number of USS10s and their poor match to the uptake motif. Note that efficient self-uptake does not necessarily imply a selective advantage, since USS accumulation in H. influenzae’s genome may simply be due to the molecular drive process.  Uptake of DNA in the respiratory track could also be influenced by the presence of chromatin and nucleoid proteins stably bound to the DNA. Although laboratory experiments typically use highly purified DNA, DNA released by cell death will be coated with these proteins, which can contribute significantly to biofilm stability (Brockman et al., 2018). Because such proteins could interfere with uptake both directly, by blocking binding to the USS, and indirectly, by blocking sliding of non-specifically bound uptake machinery along the DNA, it will be important to re-examine DNA uptake using DNA that retains its bound proteins.  LIMITATIONS OF THE STUDY Three factors limited measurements of uptake ratios: low sequencing read coverage, contamination of recovered donor DNA with recipient DNA, and segments of sequence identity between donor and recipient.  Many reads had to be excluded at the contamination-correction step, because they were in segments that were either identical between donor and recipient or were repeated within the donor genome. Strong sequence-dependent variation in read coverage caused other positions to be excluded from analysis because they had no coverage in the control input sample.  Overall, 2.3% of the genome was excluded from analysis, and an additional 1.7% was flagged as unreliable due to low coverage. In addition, the stochastic variation at low coverage positions introduced substantial noise into the calculation of experimental uptake ratios, especially at uptake valleys. On the other hand, the model predictions for long-fragment may be more accurate than indicated by their modest correlation with the noisy measured uptake ratios.  DATA AND CODE AVAILABILITY 64 All bacterial strains are available from JCM (joshua.mell@drexelmed.edu). All fastQ files have been deposited at NCBI under BioProject PRJNA387591. The corresponding BioSamples are listed in Table S1. FastQ files were also deposited at Mendeley data (DOI:10.17632/hcxp9d4zkf.1. Available at https://data.mendeley.com/datasets/hcxp9d4zkf/1). The PacBio-sequenced PittGG genome reference was deposited into Genbank under SRA number SRR10207558. Full calculations, and R scripts are available at: https://github.com/mamora/DNA_uptake.   65 CHAPTER THREE Analysis of DNA uptake bias in Acinetobacter baylyi, a bacterium without uptake specificity  Introduction DNA uptake refers to the ability of bacteria to take up DNA from the environment. It is part of a conserved process called natural transformation, which has been reported in cultured representatives of many bacterial phyla (Johnston et al., 2014) (Figure 1.2). During this process, DNA is taken up and translocated to the cytoplasm where it is either degraded or recombined into the genome.  The DNA uptake machinery is quite conserved between different bacteria (Chen and Dubnau, 2004). The majority of naturally transformable bacteria can take up DNA from related and unrelated species with similar efficiency (Carlson et al., 1983; Schwarzenlander et al., 2009). However, members of Pasteurellaceae and Neisseriaceae can preferentially take up DNA from closely related species (“self-specificity”) (Danner et al., 1980; Goodman and Scocca, 1988; Scocca et al., 1974). Self-specificity is a consequence of the uptake machinery receptor binding stronger to specific uptake sequences than to random sequences (“uptake biases”) (Berry et al., 2013).  Uptake sequences are enriched in the genome of species with self-specificity. Danner et al. (1980) proposed that uptake sequences gradually accumulated in the genome by cycles of mutation, DNA uptake bias, and recombination. Maughan et al. (Maughan et al., 2010) tested this by building a model (“molecular drive model”) to test if the accumulation of uptake sequences was possible by Danner and collaborators’ proposed mechanism.  According to this model, uptake sequences evolved by mutations increasing or decreasing, their chances to bind to the receptor. Then, uptake sequences that bind stronger to the receptor are preferentially taken up and recombined, increasing gradually in frequency in the genome (Maughan et al., 66 2010). Accumulation is expected to continue until uptake sequences gained by recombination are balanced by the ones lost by random mutations (9). This model was able to produce a frequency of uptake sequences similar to the one encountered in genomes of species with self-specificity, explaining how the evolution of self-specificity is possible independently of selection for any benefits of recombination (9).  In the genomes of species without self-specificity, no overrepresented sequences consistent with uptake sequences have been described, which has been used as evidence to suggest that species without self-specificity do not have uptake biases (Suckow et al., 2011). However, no studies have attempted to determine with precision if the uptake receptors in species without self-specificity have biases for binding to specific sequences. Mell and Redfield (2014) suggested that uptake biases could be present in species without self-specificity if these biases do not lead to a gradual increase in the preferred sequences in the genome. This would be possible if con-specific DNA is rare and does not recombine often enough, allowing random mutation to disrupt rare uptake sequences gained by recombination before they accumulate in the genome. Additionally, if the preferred sequences are simple enough to be found just by chance in most DNA fragments, then uptake sequences will not accumulate since most fragments available will already have at least one uptake sequence (Mell and Redfield, 2014). An exception would be if the uptake increases significantly as the number of preferred sequences in the fragment increases.   To gain a better understanding of why self-specificity is found in only a few species, the hypothesis that uptake biases are universally found in all transformable species as a consequence of protein-DNA interactions was tested. I predicted that physical constraints that the uptake machinery needs to overcome to pass a DNA molecule through the outer membrane pore, such as the charge and stiffness of the DNA molecule, will require some form of uptake bias. If strong protein-DNA interactions are required for DNA uptake, then uptake biases should be universally found in all transformable species. In species without self-specificity in DNA uptake, these biases are expected to be very simple (i.e. a 3-6 bp motif expected to be by chance present > 900 times in a 2 million bp genome). This study searched for uptake bias in fragments taken-up by Acinetobacter baylyi, a competent bacterium without self-specificity in DNA uptake. 67 Absence of preferred sequences in DNA taken-up by A. baylyi will weaken the hypothesis that uptake biases evolved to overcome physical limitations of passing the DNA molecule through the outer membrane.  To test my prediction, I designed an experimental platform to test for uptake bias in natural competent species. This platform consisted of recovering 200 bp synthetic fragments with a 69 bp degenerate region that had been taken up by competent cells. Recovered fragments were then compared with the sequences of fragments before uptake to detect uptake preferences for short sequence motifs. Acinetobacter baylyi was chosen as a model organism to be tested with this platform since it lacks self-specificity, competence can be easily induced, and it can achieve high frequency of transformation.  A. baylyi, previously known as A. calcoaceticus, is a gamma-proteobacteria from the Moraxellaceae family. It can be found in soil but it has been reported to act as an opportunistic pathogen in immunocompromised patients (Li et al., 2015; Palmen et al., 1993). Competence can be easily induced by diluting overnight cultures in fresh media, obtaining transformation frequencies up to 5x10-2 (12). A. baylyi is able to take up plasmid and chromosomal DNA from related and unrelated species (Palmen et al., 1993, 1994). This ability to take up chromosomal DNA from unrelated species indiscriminately, make A. baylyi an ideal candidate species for our study, since finding a motif enriched in recovered sequences from competent A. baylyi cells would confirm the hypothesis that uptake biases are required for DNA uptake; while the absence of a motif will suggest that uptake biases are not universal in all competent species and are not a requirement to overcome the physical limitations of the uptake process.  Materials and methods Strains and growth conditions For the DNA uptake experiments, A. baylyi wild type strain BD413 and T205 (ADP239 comP::nptII) were used (Drotschmann et al., 1997). Strains were grown in Luria broth (LB) media or Luria broth agar at 30 ℃.  T205 had a knockout mutation in comP, which is a gene required for DNA uptake (Drotschmann et al., 1997). This strain was used as a no-uptake negative control in the experiments.  68 Natural transformation A. baylyi was made competent by diluting 0.5ml of an overnight culture growing at 30 °C into 30ml of LB broth (Palmen et al., 1993). Cells were incubated with shaking for 1 hour. A 0.5 ml aliquot was taken and was incubated with 1 µg of Kanamycin resistant DNA from T205 strain for 60 minutes. 10 µl of DNase I (100 µg/ml) were added and incubation continued for 5 min at 30°C. Cells were then incubated for 60 more minutes and then were plated on LB media plates with and without 15 µg/ml of kanamycin.  DNA from T205 was obtained following a phenol-chloroform DNA extraction as described by Sambrook (2001). Briefly, 500ul of cells were incubated with 0.1 mg of lysozyme for 5 minutes, followed by 50 µg of RNAse and 0.2 mg of proteinase K and 10 µl of SDS 10%. One volume of phenol:chloroform (1:1) was added to the cells. The mixture was centrifuged, and the aqueous phase recovered. This step was repeated adding half a volume of chloroform instead of phenol:chloroform in the mixture. DNA was then precipitated after adding 900 μl of ethanol 95% and 15 μl of 5M sodium acetate. The pellet was washed with 500 μl of 80% ethanol and resuspended in in 500 µl of ddH2O. Synthesis of double-stranded input DNA fragment Synthetic dsDNA fragments were built by annealing and extending two synthetic long PAGE-purified ultramerÒ oligonucleotides (Integrated DNA technologies). Both oligonucleotides were designed with Nextera XT sequencing primer binding sites (Figure 3.1) (required copyright notice: oligonucleotide sequences © 2018Illumina, Inc. All rights reserved. Derivative works created by Illumina customers are authorized for use with Illumina instruments and products only. All other uses are strictly prohibited).  The left oligo contains: a 33 bp primer binding site (red), 3 bp of fixed bases, a 69 bp fully degenerate region (with 326,276 different unique sequences sequenced), and a 30 bp spacer complementary to the right oligonucleotide (Figure 3.1). The right primer contains: a 34 bp primer binding site (purple), a 31 bp fixed nucleotide region, and a 30 bp spacer complementary to the left oligonucleotide (Figure 3.1).  To generate a 200 bp synthetic fragment, both oligos were annealed by mixing 50pmol of each and then heating the mix in a dry-heat block for 5 minutes at 95 °C. Then the oligo mix was left 69 to gradually cool with the heat block off, until it reached 30 °C. Five microliters of annealed-oligos were then extended with 1U of Thermo taq polymerase (NEB), 1X Thermo buffer (NEB), and 300 µM dNTPs. The resulting synthetic fragments were purified using Econospin ® columns (Epoch life sciences) and were quantified using a Qubit dsDNA HS Assay Kit (ThermoFisher).   Figure 3.1: Diagram of synthetic fragments used for the uptake experiments. First, two oligos were annealed (A) and extended (B). Each region of the fragment is color coded and the sequence of each oligo is shown in (C). Red and purple are the Illumina sequencing primer binding region. Black are the 3 fixed bases that were used to measure synthesis errors (see Methods). Green are the 31 bp fixed bases region. Orange shows the annealing area between the two oligos.  In grey, the 69 bp degenerate bases region is shown.  DNA uptake assays DNA uptake experiments used wild type strain BD413 as recipient, using synthetic fragments as input donor DNA. Three replicate experiments were done using the wild type strain. Additionally, as a no-uptake control, strain T205 was used as recipient.   Acinetobacter strains were made competent following the protocol described in natural transformation section of the Results. Ten ml of competent cells were incubated, under conditions described above, with 0.9 µg of synthetic fragment DNA for 30 minutes. 10 µl of DNase I (100 µg/ml) were added and incubation continued for 5 min. Cells were washed by 3 Left oligo:Right oligo5’TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGGTANNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNGTGCCACCTGATGCGGTGTGAAATACCGCA3’5’GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGCCGCATCATGTGCTCAGAGGCTACAGTAGAGTGCGGTATTTCACACCGCATCAGGTGGCAC3’C.69 bp33 bp 3 bp 30 bp 34 bp31 bp5’5’3’3’5’3’3’5’A.B.70 centrifugation steps at 4000 x g for 15min, followed by a resuspension step with 1 ml minimal media (prepared as by Palmen et al. 1993), except for the last centrifugation step in which cells were resuspended in T50E50 buffer (50 mM Tris, 50 mM EDTA pH 7.5). DNA was then recovered by phenol-chloroform extraction and ethanol-precipitation as described above, with the exception that final elution was in 40 µl of ddH2O.   Recovered DNA was amplified by PCR.  Conditions include: 400 nM of each primer, complementary to the first bases of the synthetic fragment (Fw: TCGTCGGCAGCGTCAGATGTGT, Rv: GTCTCGTGGGCTCGGAGATGTGTAT), 200 µM dNTPs, and 1U Thermo taq polymerase (NEB). PCR program consisted of a 95 °C denaturation step (5 min) followed by 25 cycles of a 95 °C denaturation (30 sec), 58 °C annealing (40 sec), 68 °C extension (30 sec), and a final 68 °C extension (10 min). Amplicons were purified using AMPure XP beads before the library preparation step. Sequencing and library preparation Libraries were prepared following 16S sequencing library preparation guide (Illumina, 2013), with the exception that the 1st round sequencing primer attachment PCR step was not needed since the synthetic fragments already contained the sequencing primer regions. Samples were sequenced by the UBC Sequencing and Bioinformatic consortium using an Illumina Miseq micro run with 2x250nt for each library. Donor input synthetic fragments were sequenced as a control to test enrichments of motifs taken up by the wild type strains. Bioinformatics Sequences were extracted out of fastq files. For the analysis of the error and efficiency, the first 61bp of the right-end of the sequencing primer was extracted. All other subsequent analysis used the randomized sequences from left and right ends of the sequencing primer. To determine the overall effect of PCR cycles on fragment recovery, I assessed the number of unique sequences vs repeated sequences in the randomized 69 bp region of the synthetic fragments sequenced in each sample, before and after recovery. This was done using scripts in R v3.5.1. Only sequences with unique randomized regions were used, discarding duplicated reads (Table 3.1). This step removed reads containing errors during the PCR and library-preparation steps. 71 Sequences from left and right ends of the sequencing primer  were merged into one file per dataset.   Presence of uptake bias was tested by dividing the randomized region of the recovered sequences into kmers, where k was equal to 4, 5, 6 or 7 nucleotides, using jellyfish, v2.3.0 (Marçais and Kingsford, 2011). Kmer counts were normalized by dividing their number by the total number of kmers. Sequence enrichment was determined by comparing normalized kmer frequency of input and recovered DNA. To assess the impact of synthesis errors before and after recovery, I examined the normalized kmer frequency of the non-degenerate first 50 sequenced base pairs starting at the right end of the fragment (Figure 3.1, green and a fraction of the orange section). To test how often incorrect bases were introduced by either the sequencing or the amplification steps, error rate frequency was calculated by assessing the frequency of mismatches of the first 3 fixed nucleotides bases (Figure 3.1, black section) in the input and in the recovered samples. Presence of enrichment motifs from each dataset were analyzed using MEME v.5.0.3 and using Selex v3.8 R package (Bailey and Elkan, 1994). Statistical significance of enriched kmers frequency from recovered datasets were compared statistically with input kmers frequency using a chi-square test from the chisq.test function of the ‘stats’ R-package. Probability of finding a sequence motif in the 69 bp degenerate region and 131 bp fixed region of the fragments was calculated as the binomial probability of finding a random 5-7 bp sequence in a 69 and 131 bp fragment, in both orientations, using the ‘dbinom’ function from the R-package ‘stats’. This was done using base frequencies matching the randomized region of the input fragment (see the ‘Sequence enrichment’ section below).  Sensitivity of each analysis was determined by replacing 0.67% to 13.5% of random 5 bp positions of the degenerate region from the input fragment reads with the sequences “CGAAC”. Statistical differences between the 5mers from the input fragment reads and from the reads with the different percentages of the enriched sequences were tested using chi-square as detailed before. 72 Motif finder analysis Motifs enriched in recovered datasets were searched for using MEME-suite v.5.0.3 using the algorithms MEME and DREME.  MEME algorithm was used to find motifs enriched in the recovered datasets compared to the input dataset as control. DREME algorithm was used to search for short (less than 8bp) ungapped motifs. To determine the sensitivity of MEME and DREME, I replaced 13.5% (2-fold), 20.2% (3-fold), and 26.9% (4-fold) of random 8 bp positions of the degenerate region from the input fragment reads with the partially degenerate sequence ‘AASSNSST’, where S is C/G (50:50) and N is A/T/C/G (25:25:25:25). MEME was used to search for motifs with 6– 30 bp using the 0-order background model to account for base frequency.  The differential enrichment mode was used to find relative enrichment of sequence motifs relative to the input degenerate region, as a control. DREME algorithm was used to search for motifs with 4 - 8 bp, using the input dataset as a control.    Results and Discussion Recovery of DNA To detect bias in A. baylyi DNA uptake, 900 ng of 200 bp synthetic fragments were incubated with competent A. baylyi recipient strains BD413. To avoid sequencing contaminating chromosomal DNA, the input synthetic fragments were designed with flanking sequencing primer binding sites (Figure 3.1). These sites also allowed me to recover the fragments by PCR. I used competent cells isolated from 3 independent colonies as well as cells from a mutant strain without detectable DNA-uptake (Table 3.1). These uptake-negative recipient strain helped to determine the effects of recovering input DNA that was not eliminated during the wash steps (explained below). Table	3.1.	Number	of	sequenced	reads	for	each	sample,	as	well	as	the	percentage	of	reads	with	unique	degenerate	regions		73  Following the incubation step, I added a DNase I step and 3 washes to degrade and wash away input DNA fragments to prevent them from being amplified. DNA taken-up was recovered by phenol-chloroform genome extraction and amplified by PCR, with 25 cycles.  PCR artefacts and quality controls  Errors in the sequencing and amplification steps could be a confounding factor in analysis searching for biases in DNA uptake. To determine the frequency of errors in the amplification steps, I analyzed the percentage of undefined (i.e. a N instead of a G) or incorrectly called bases (i.e. a T instead of a G) on the 61 fixed base pairs from all reads, starting by the right-end (green and orange section in Figure 3.1B). Assuming no sequencing errors, fixed bases should be the same in every read, so the percentage of incorrect bases could be calculated by comparing reads sequences to the synthetic fragment sequence. Input fragments showed a small percentage (2.15%) of duplicated reads, presumably as a result of errors during cluster amplification at the library preparation synthesis step. Sequencing error due to incorrect base calling on all datasets was between 0.87- 0.94% ranging from 0.1 – 2% across all 61 positions. The wide range of error across positions was suggestive of reproducible sequencing biases which increased with each base pair synthesis step (Figure 3.2A).  This was confirmed by the very small standard deviation of error observed across all positions, which ranged from 0.0001 – 0.0009 (Table S3). On the other hand, ten positions showed undefined bases (bases with ‘N’ instead of A,C,G, or T). The second position showed the highest percentage of undefined bases ranging from 54-74 % of the reads among datasets (Figure 3.2B). The other 9 positions showing undefined bases had, on average, frequencies ranging from 11-21% among datasets. Seven out of ten positions with Strain genotypetotal number of read pairsnumber of reads used in the kmer analysispercentage of reads with unique degenerate regionsBD413 replicate 1 wild type 413524 406195 68.77BD413 replicate 2 wild type 323105 318985 73.96BD413 replicate 3 wild type 352796 348311 88.95T205 (no uptake control) pobA comP::nptII  KmR 383055 377225 16.56input fragments 333445 330618 97.8574 undefined bases were among the first fifteen bases sequenced, suggesting that most of these sequencing errors occur during the first sequencing cycles (Figure 3.2B).   Figure 3.2. Incorrect base calling analysis. A. the frequency incorrect base calling at a 61 bp section of the fixed region of the fragments of the input (red) , 3 replicates (green, blue and purple) and the no-uptake control (olive) dataset. B. the frequency of undefined bases, called as 'N', among all datasets.  Since the first position of reads showed higher number of undefined bases, for subsequent analysis I used only reads that matched correctly the first 3 bases of the fixed sequence region (black region, Figure 3.1A). This was a control step to select reads that had a good quality, since I could not assess read quality by aligning the reads to a reference sequence. Approximately 2% of the reads per sample were eliminated during this control step (Table 3.1). 300k to 400k pairs of reads were sequenced per sample.  Each input fragment sequence was expected to be unique, given the length of fragment degeneracy region (69 bp) and the lack of exposure to multiple PCR cycles. A low number of recovered fragments would reduce the diversity of sequences and limit the amount of template CCGCATCAGGTGGCACGAGGATGCGGAAGAGTGCGGTATTTCACACCGCATCAGGTGGCAC0 20 40 60A.B.75 available for the first rounds of PCR amplification.  As a consequence of this, samples with poor recovery were expected to have a high percentage of repeated sequences after PCR amplification. The number of unique sequences (singletons) were measured to assess the quality of recovered samples. Amplified DNA from the no-uptake control was expected to have a high percentage of repeated sequences since all the DNA came from residual contaminant input fragments; while amplified recovered DNA from the wild type strain replicates was expected to have a low number of repeated sequences. As expected, sequences from no-uptake recipient strain contained higher percentage of repeated sequences (83.44%) (Table 3.1) compared with the input fragments (2.15%). This showed that the no-uptake controls had a very low number of fragments. The three Acinetobacter BD413 replicates had less repeated sequences than singletons (11.05 - 31.23%), suggesting that most of their recovered amplified sequences came from a high number of starting sequences and not only from residual input DNA.  Sequence enrichment To detect uptake bias, my strategy relied on determining if there were sequences overrepresented in the DNA taken up by competent A. baylyi cells. This was done by using an analysis of kmer frequencies on recovered vs input datasets. Finding overrepresented kmers on the recovered DNA datasets, compared to the input dataset would be a strong indication of uptake bias in A. baylyi. As a positive control, I examined the ability of this technique to statistically differentiate sequences found more often in the recovered samples than in the input controls (sequence enrichment). This was done by analyzing the degenerate region of the input reads, with and without artificially replacing different percentages of random sections of reads with a particular 5mer sequence and its reverse complement. I determined the minimum increase in the frequency needed to statistically detect the differences in 5mer frequencies. The input degenerate region was synthesize using equal proportion of all 4 bases. However, after synthesis and PAGE purification, we detected an uneven base frequency with 34% T, 28% G, 23% A and 16% C (44% GC content). Uneven base frequencies affected the kmer frequency since I observed uneven 5mer frequencies, ranging from 0.0005 for 5mers rich in the least frequent base ‘C’ to frequencies of 0.0031 for 5mers rich in the most frequent base ‘T’ (expected frequency with 50% GC was 0.001).  76 To test if the uneven kmer distribution affected the ability to detect sequences enriched in the recovered fragments, I replaced random 5mers in the input dataset with known 5mers that were in high (TTTTT = 0.003), medium (GACTA = 0.0009) and low (GCGCC = 0.0005) expected frequency. This was done to increase the frequency of the known 5mers in the input dataset until they were statistically different from the original kmer frequency (0.02 to 0.3-fold increase in frequency) (Figure 3.2A). Low and medium frequency 5mers, that were artificially enriched by more than 0.15-fold, were statistically different from the original input dataset kmers (Figure 3.2B and C). High-frequency 5mers became statistically significant when artificially enriched at 0.09-fold (Figure 3.2D). This analysis suggests that even very small differences in 5mers frequencies will be detected by the analysis.      Figure 3.2 Comparison between the frequency of artificially enriched 5mers in the input dataset, with low, medium and high frequency, with their original kmer frequencies. A. distribution of the input 5mers. Red arrows showed the frequency of the 5mers chosen to be A.FrequencyCountsFold enrichmentB.P-valuelow frequency 5mer medium frequency 5mer high frequency 5merD.C.Fold enrichment Fold enrichmentB.C.D.2040608000.001 0.002 0.0030.1 0.2 0.3 0.1 0.2 0.3 0.1 0.2 0.300.250.500.751.0077 artificially enriched. B-D. P-value of chi-square analysis of 5mers with low (B), medium (C) and high (D) frequency with and without having their frequency enriched by 0.02 to 0.3-fold (x-axis).   Regardless of kmer size analyzed (k = 4-7), no significant differences (P = 1) were observed between the recovered fragments and input datasets in any of the 3 independent replicates. However, significant differences were observed between the no-uptake control and the input dataset (P = <2e-16). Figure 3.3 A-D shows normalized 5mer frequency of each Acinetobacter replicate sample divided by the frequency of input 5mers. Variance in the differences between 5mers in the recovered dataset by input datasets were greater at lower frequencies. However, these differences were not greater than 10%, except on 5mers from the no-uptake control. These results suggest that there is no detectable uptake bias in Acinetobacter samples. Statistical differences observed between the no-uptake control and the input dataset are probably a consequence of the low percentage of unique reads, which is expected to shift the frequency of the kmers in duplicated reads (from 0.0001 - 0.005 in the no-uptake control dataset vs 0.0005 – 0.003 in the input dataset ) (Table 3.1).  78    Figure 3.3: Normalized frequency of 5mers in the 69 bp degenerate region from the recovered (y-axis) divided by the input (x-axis) fragments. 5mers from the Acinetobacter BD413 recovered replicates are shown in A-C, while 5mers from Acinetobacter no-uptake control (T205) is shown in D.  MEME analysis Motifs with several degenerate bases could be difficult to detect using traditional kmer analysis. I used two algorithms from the MEME-suite v. 5.0.1 to test for uptake preferences for motifs with degenerate bases.  To test if these two algorithms, MEME and DREME could really detect complex motifs with degenerate bases, I replaced random 8mers in the input dataset with 8mers belonging to the sequence motif in Figure 3.4A. The MEME algorithm was able to detect motifs enriched 2-fold or more but failed at finding the motifs enriched less than 2-fold (Figure 3.4B). In contrast, DREME A. B.C. D.Normalized frequency of replicate 1Frequency of recovered / input fragmentsNormalized frequency of replicate 2Normalized frequency of replicate 3 Normalized frequency of no-uptake controlFrequency of recovered / input fragmentsFrequency of recovered / input fragmentsFrequency of recovered / input fragments79 algorithm, which is efficient at finding short-sequence motifs (4-8 bp), was able to detect the motif even if this was enriched only 1.5-fold (Figure 3.4B).     Figure 3.4 Control sequence-motif used to determine sensitivity of MEME and DREME algorithms. A. sequence motif used to replace random 8mers of the input dataset to create a positive control dataset. B. Table showing the sensitivity of MEME and DREME algorithms when these algorithms were used to detect the Motif in A. when this was replaced in 10.11 to 26.95% of the input reads. E-value, as described by MEME-suite, is equal to the P-value calculated using Fisher exact test times the number of candidate motifs tested.   In the three replicate recovered fragment preparations from Acinetobacter competent cells, MEME algorithm did not find any motif. In contrast, DREME algorithm was able to find motifs significantly enriched compared to the input fragments (Figure 3.5). However, none of the motifs found were consistently present among the three replicates of the recovered fragments from Acinetobacter competent cells. This suggests that our MEME-analysis provided no evidence of uptake bias among Acinetobacter competent cells.   Fold E-value DREME E-value MEME %_of_reads_replaced with the motif4-fold 2.1e-20791 1.5e-12956 26.95%3-fold 6.3e-13101 1.1e-8178 20.20%2-fold 2.0e-6698 5.0e-2070 13.48%1.5-fold 4.2e-4099 not detected 10.11%A.B.80      Figure 3.5: Analysis of the 3 motifs with highest E-values in recovered fragments replicates and the no-uptake control with the DREME algorithms. Conclusion The lack of enrichment of motifs in the analysis suggests that either the sequence motif that is preferred by the uptake receptor is already present in the fixed regions of the synthetic fragment, or that sequence-dependent uptake bias is very weak or not present at all. The probability of there being a sequence promoting uptake in the fixed region of the synthetic fragments depends on the uptake-sequence length and on the length of the fixed region (131 bp in total). If such a sequence existed (3-7 bp, not degenerate), Table 3.2 shows the probability of encountering at least one occurrence of it in the fixed and degenerate regions of the fragments, using the binomial probability distribution and taking into account the GC content of the degenerate region of the sequenced synthetic fragments. Since the base composition of the sequence that might be enriched in the recovered dataset is unknown, for simplicity I assumed that the probability of having any of the four nucleotides was 0.25. The impact of the bias in the base composition of the degenerate region will be addressed below.   No-uptake controlReplicate 1 Replicate 2 Replicate 381 A sequence motif shorter than 5 base pairs is expected to be present in most of the fragments, so a preference for it would not be detected by the analysis since most fragments will be taken up at similar efficiently. In this case, uptake bias would be detected only if fragments with 2 or more preferred sequence motifs have much higher uptake than fragments having only one. This is unlikely since results from Ambur et al. (18) in Neisseria and the chapter 2 results in Haemophilus influenzae suggests that short fragments (0.2-0.6kb) with 2 or more uptake sequences do not have uptake higher than fragments with 1 uptake sequence.  Table	3.2.	Probability	in	percentages	of	having	at	least	one	of	a	3-8	bp	long	sequence	in	the	69	bp	degenerate	region	and	131	bp	fixed	region	of	the	fragment.	Percentages	were	calculated,	based	on	the	GC	content	of	synthetic	fragments,	for	kmers	with	A	or	Ts	as	well	as	C	or	Gs.		 Given the bias in the base composition of the oligos degenerate sequence, motifs rich in A or T will have higher chances to be present in the fragments than motifs rich in G or C. For instance, taking in account this bias base frequency, I estimate that 29.72% of the degenerate regions of fragments will have at least one TTTTT, while only 11.3% will have at least one GGGGG. If the motif is 6 bases, the probability falls to 10.4% when having at least one TTTTTT, and to 3.1% when having at least one GGGGGG in the degenerate region. This suggests that motifs with 5 or fewer bases that are rich in A or Ts won’t be detected since they will be present at least once in more than a quarter of all fragments. Such short sequences rich in A or T have been reported in DNA-binding sequences that have a specific DNA shape, such as rigidly bent A-tracts at nucleosome binding sites or narrow minor grooves at AT-rich transcription factor binding sites (Abe et al., 2015; Rohs et al., 2009; Stella et al., 2010; Yella et al., 2018; Zhurkin et al., 2005). Since several different sequences could Kmers A or T C or G A or T C or G3mer 100% 100% 100% 100%4mer 75.7% 37.7% 100% 62.6%5mer 29.7% 11.3% 51.8% 21%6mer 10.4% 3.1% 19.5% 6.1%7mer 3.5% 0.9% 6.8% 1.7%8mer 1.1% 0.2% 2.3% 0.5%69bp degenerate region 131 bp fixed region82 promote a similar shape, uptake bias for DNA shape features would be harder to detect. According to Duzdevich et al. (24), all DNA-binding proteins have some degree of sequence-specific, as well as, DNA shape-specific DNA binding. Sequence-specific binding or base-readout has been defined as binding interactions between the DNA bases and the protein amino acids; while shape-specific binding or shape-readout has been defined as interactions between protein amino acids and the DNA backbone (Rohs et al., 2010). In absence of strong sequence-specific bias, it is possible that there is very weak bias caused by a shape-specific binding mechanism acting on Acinetobacter DNA uptake.  In summary, uptake bias was not detected in A. baylyi. This may be because the kmer analysis could not detect motifs smaller than 5 bp. This suggests that uptake biases in A. baylyi are either absent or limited to a very short sequence. Absence of a detectable preferred sequence motif in the recovered fragments suggests that uptake bias isn’t a universal process. It is still unclear why strong uptake bias have evolved in only a limited number of species, while others, such as Acinetobacter, have no detectable uptake bias. It has been suggested that these differences are a consequence of uptake-biases evolving adaptively as a mate-recognition mechanism in species with self-specificity. However, considering that the donor DNA comes from a dead bacteria, this is unlikely since evolving natural transformation for the benefits of genetic variation by recombination would require either an extreme form of altruism, or exceptional circumstances such as live cell being very unfit  (Redfield, 1988). Even then, DNA from dead unfit cells will remain available to be able to recombine lowering the fitness of other more fit cells taking up their DNA. These forms of altruism are rare and only seen when group selection is stronger than individual selection (Wilson, 1975). Additionally, it has been shown that the main fitness advantage of DNA uptake of competent vs non-competent cells from  A. baylyi is recombination-independent (Hülter et al., 2017). Still, our results suggest that uptake bias are not a requirement of the uptake machinery to overcome physical constraints of getting a long relatively stiff DNA molecule through the tight outer membrane pore. Persistence length of DNA molecules depend on their size, so a limitation of this study was that only 200bp fragments were considered. It is possible that DNA molecule length has an impact on uptake. Additionally, DNA-binding receptor has not been identified in most naturally competent species. It is very clear that this protein is 83 not conserved and likely evolved independently among naturally transformable species. The next step to truly understand why uptake biases are present in only a limited number of species is to determine how exactly each DNA-binding protein binds strong enough the DNA without any detectable specific binding in species such as A. baylyi.                84    CHAPTER FOUR Conclusions This project focuses on understanding uptake bias in bacterial species with and without self-specificity. The goal was to fully characterize factors that have an impact on uptake bias in one species with self-specificity, choosing Haemophilus influenzae as a model, as well as to determine if uptake bias is present in a species without self-specificity, choosing Acinetobacter baylyi as a model. Both of these goals have evolutionary implications in understanding the evolution of uptake bias. On the one hand, uptake biases could influence the recombination of DNA fragments across the genome, which could help to understand the role of uptake biases in genome evolution. On the other, determining if uptake biases are present in species without self-specificity could help explain how they have evolved and if these biases are a consequence of physical constraints of passing a long DNA molecule through the outer membrane pore. In the second chapter, I measured DNA uptake at every genome position in H. influenzae and compare its results with predicted uptake generated by a computer model. This study was able to determine the effects of USS-dependent factors that are relevant to uptake, including the positional dependencies between different parts of the uptake motif, the interaction effects of bases within the USS motif, and the effect of multiple USSs per DNA fragment.  In the third chapter, the presence of overrepresented sequences in taken up DNA from A.baylyi were examined. No evidence for overrepresented sequences was found in the analysis, suggesting that uptake biases are not universal in competent species. These results weaken the hypothesis that strong uptake biases are required to overcome physical constraints of moving the DNA molecule through the outer membrane.  85 Evolution of uptake bias and molecular drive Among all naturally transformable species only a few, such as members of Pasteurellaceae and Neisseriaceae, preferentially take up DNA from close relatives. These preferences have been associated with uptake sequences that are enriched in their genomes. Danner and collaborators (Danner et al., 1980) proposed that gradual accumulation of uptake sequences, due to cycles of DNA uptake bias and recombination of fragments with uptake sequences, were responsible for these preferences in uptake of conspecific DNA. Maughan and collaborators (Maughan et al., 2010) tested this hypothesis using a molecular drive model to test if mutation and recombination alone could explain the frequency and distribution of uptake sequences. This molecular drive model assumes that the uptake machinery receptor has increase affinity to a particular DNA sequence, making uptake more efficient. Our two study organisms can be considered as two extremes with respect to the affinity of the uptake receptor for DNA sequences. A. baylyi did not have any detectable bias, while H. influenzae showed very strong uptake bias with up to a 1000-fold difference in uptake of segments with and without a USS. Strong biases, according to Maughan et al (3), will promote uptake sequences with a higher affinity to the receptor to be enriched in the genome. This is consistent with the high average of USS scores (11.5 bits for NP and GG) observed in Haemophilus NP and GG genomes. The analysis of Acinetobacter DNA uptake suggests that uptake biases, if present, have to be very weak, less than 0.1-fold (if the sequence is a 5mer, Figure 3.2). This suggests the uptake receptor does not necessarily need to bind with high sequence specificity during DNA uptake.  Weak or no biases in the DNA uptake process contradicts our original hypothesis that uptake biases are a requirement to overcome the physical constraints of pulling the DNA molecule through the outer membrane pore.  This analysis does not imply that the uptake receptor will bind to all sequences with equal efficiency or that uptake biases evolved as mate-recognition mechanism. It remains unclear how exactly the uptake receptor binds strong enough to the DNA in A. baylyi without detectable uptake biases. Given that the uptake receptor does not seems to be conserved among naturally transformable species, it is likely that this mechanism differs slightly among species. In general, DNA binding proteins associate to the DNA by weak 86 interactions that allow them to slide through the DNA until they find a specific sequence that interacts strongly with the protein. These interactions could be by hydrogen bonds between amino acids and the bases, or by electrostatic interactions of positively charged amino acids and the negatively charged backbone phosphates (Cherstvy, 2009; Harteis and Schneider, 2014; Holbrook et al., 2001; Redding and Greene, 2013). Uptake machinery in Acinetobacter could be binding specifically to a simple but highly degenerate sequence motif that induces DNA shape features that increase the binding strength (6). DNA shape features can allow for specific hydrogen bonds between the protein side chains and the DNA bases that would not be possible if the DNA had other shape features (Cherstvy, 2009; Gordân et al., 2013; Rohs et al., 2009). These interactions have been described in many DNA-binding proteins that were reported to lack strong binding specificity (Harteis and Schneider, 2014; Subramani et al., 2010; Wei et al., 2014). As proposed by Mell and Redfield (Mell and Redfield, 2014), such simple motifs might be found frequently in most genomes, in which case molecular drive will not promote gradual enrichment of the uptake sequences, explaining the absence of uptake preferences in most naturally transformable bacteria. For self-specificity to evolve by molecular drive, competent bacteria have to live in environments with abundant DNA of their own species. As a respiratory tract pathogen, H. influenzae is able to form biofilms in which conspecific DNA is present, especially during infections. In this environment, H. influenzae DNA must compete for uptake with abundant human DNA. Our simulation analysis showed that Haemophilus DNA is able to outcompete human DNA for uptake, even when this is 100-fold in excess. This is a consequence of the molecular drive acting on the Haemophilus genome, and not in the human genome, promoting high scores and frequency of USSs.  In contrast, A. baylyi, a gram-negative bacteria belonging to a wider class of Gammaproteobacteria, is usually found in soil (Atrouni et al., 2016) and only very rarely reported as an opportunistic pathogen in humans (Li et al., 2015). Even though these bacteria are able to become competent and take up DNA in biofilms (Merod, 2014), its DNA has to compete for uptake with a huge diversity of other species’ DNA. Soil is a very complex environment, and it can have more than 2 x 109 prokaryote cells and 2000 – 18000 different genomes in one gram 87 (Daniel, 2005). The absence of uptake biases in Acinetobacter suggests that its DNA will be competing very poorly for uptake with the abundant DNA from other organisms.   The molecular drive model also requires that a relatively large fraction of the DNA taken-up be recombined. Both Haemophilus and Acinetobacter have been reported to efficiently recombine fragments larger than 1 - 3kb (Overballe-Petersen et al., 2013; Pifer and Smith, 1985). Strong uptake bias in H. influenzae allows self-DNA to outcompete the abundant host DNA; however, that is not the case of A. baylyi. Competing poorly with DNA from distantly related species might limit the recombination of these bacteria to biofilms communities, if the DNA from their own species is abundant enough (Merod, 2014).  Recombination has been frequently reported in H. influenzae from otitis media isolates with a recombination/mutation rate of 3.7 (Vos & Diderot 2009, Cody et al 2003, Meats et al 2003). It is not clear how often recombination events occur in A. baylyi, but a smaller fraction of recombination/mutation is expected based on the 1.3 recombination/mutation rate reported in A. baumannii, another naturally transformable Acinetobacter species. Still, 526 recombination events were reported in the genome of A. baumannii with an average fragments size of 2.1 kb long (Touchon et al 2014).  This suggests that despite Acinetobacter DNA being expected to compete poorly with DNA from distantly related species, recombination might still happen by other horizontal gene transfer mechanisms or by transformation when its own DNA is abundant enough.   Molecular drive predicts that uptake sequences distribution depends mainly on the length of the fragments taken up and will be limited by selection against disrupting genes function (Maughan et al., 2010). This is consistent with Findlay and Redfield’s (Findlay and Redfield, 2009) analysis showing that USS and DUS are found less often in housekeeping genes and that their distribution was not correlated with gene functions. Additionally, H. influenzae uptake ratios across the genome, using the long fragments dataset,  showed to be within two-fold of the mean uptake for 90% of the genome. This is consistent with most long fragments taken-up having at least 1 USS, given the average density of uptake sequences in Haemophilus being 1 every kb (Smith et al., 1995), similar to Neisseria (Smith et al., 1999). This is expected given the relatively large size (300 bp to 50kb ) of extracellular DNA fragments in the respiratory tract (Lethem et al., 1990).  88 Next steps, what remains to be understood The absence of detectable overrepresented sequences in recovered DNA from A.baylyi suggests that, if uptake biases are present, they have to be weak.  This could be possible if the uptake receptor binds specifically to simple short sequences that are found by chance often in most genomes.  Retraction of the uptake pilus is able to generate great forces needed to pull the DNA through the outer membrane pore (Maier et al., 2002).  One question that these findings raise is how the receptor is able to bind strong enough to the DNA if the uptake receptor binds to a very simple motif or indiscriminately to any DNA sequence. It is possible that shape-readout mechanisms increase the binding strength of the uptake receptor to simple DNA motifs. However, DNA shape features are a consequence of interactions between DNA bases, so sequences promoting a specific DNA shape features have to be very subtle to not be detected by our analysis.  Unfortunately, without identifying the uptake machinery receptor, it will be very difficult to advance any further to understand how the uptake receptor binds to the DNA. This protein has already been identified in N. meningitidis (Berry et al., 2013), so an alternative could be to use Neisseria as a model to measure the strength of binding specificity using single-molecule techniques, such as optical or magnetic tweezers. However, a problem with this strategy is that no uptake receptors have been identified for species without self-specificity, which limits the ability to measure the strength of DNA binding in species without strong uptake bias.  The impact of DNA shape on DNA uptake of H. influenzae needs also to be carefully characterized. Specifically, a greater understanding is needed of exactly how different DNA shape features, such as narrow minor groove width at the AT-tracts, alter uptake probability. This is not trivial since many DNA shape features could influence DNA binding (Li et al., 2017). Additionally, some DNA features are correlated and could interact with each other, such as minor groove width and propeller twist (Rohs et al., 2009). To do this, future experiments could be designed to recover taken up fragments, each with the same USS inner core but different flanking sequences. By recovering and sequencing these fragments, the impact of different DNA 89 shape features could be determined, independently from specific binding to the USS's inner core. Other factors that might influence DNA uptake are nucleoproteins or structural modifications associated with DNA.  DNA used in uptake and natural transformation experiments is usually purified from any proteins associated with it. However, extracellular DNA could still have nucleoproteins associated. These proteins could influence DNA uptake at the binding or uptake steps, as they have strong affinity to A-tracts or AT-rich regions, similar to the USS tracts (Holbrook et al., 2001; Shimizu et al., 1995). Structural modifications to DNA, such as methylation or acetylation, could also influence DNA uptake. For instance, ethylation of bases within a USS have been shown to influence uptake positively or negatively depending on which phosphate is modified (Danner et al., 1980). Methylated DNA have also been shown to have a positive or negative impact on DNA binding of transcription factors (Yin et al., 2017). Similarly, methylation of an overrepresented sequence in Campylobacter jejuni genome promotes its DNA uptake and transformation (Beauchamp et al., 2017). Future experiments could incorporate dam/dcm methylation into different bases of the consensus USS sequence in synthetic DNA fragments. This analysis would allow us to measure the effect of methylation in uptake of USS sequences.  In conclusion, to fully understand all factors involved in DNA uptake, interactions between the DNA-uptake receptor and DNA fragments need to be fully characterized for both A. baylyi and H. influenzae. Understanding this proteins DNA interactions would clarify the role that DNA shape plays in DNA uptake biases in H. influenzae, as well as help us understand how the DNA uptake receptor in A. baylyi could bind strong enough to DNA to overcome physical contrains of taking DNA inside the outer membrane without any detectable sequence-specific DNA-binding.     Unfortunately, in the case of DNA uptake of H. influenzae and A. baylyi this was difficult since the uptake receptor binding to DNA is unknown. Despite this limitation, my study was able to determine that DNA uptake in H. influenzae not only depends on the interactions between the uptake machinery and individual bases of the USS but on the DNA-shape features, such as narrow minor groove width and higher propeller twist at the AT-tracts. These features suggest rigid bending in these regions of the USS. Similar DNA shape preferences have been identified in 90 proteins that exert high degree bending forces on DNA molecules. For instance, HU and IHF nucleoproteins are able to bind and bend the DNA molecule without a strong sequence-specific DNA binding (Shimizu et al., 1995). This mechanism might be the key to understand why no uptake bias was detected in A. baylyi. If that is the case, I could speculate that in species without self-specificity, DNA shape features of the binding motif might be increasing the binding strength enough to overcome the physical limitations of passing the DNA molecule through the outer membrane pore.                                   91 Bibliography Abe, N., Dror, I., Yang, L., Slattery, M., Zhou, T., Bussemaker, H.J., Rohs, R., and Mann, R.S. (2015). Deconvolving the recognition of DNA shape from sequence. Cell 161, 307–318. Albritton, W.L., Setlow, J.K., Thomas, M., Sottnek, F., and Steigerwalt, A.G. (1984). Heterospecific transformation in the genus Haemophilus. Mol. Gen. Genet. 193, 358–363. Ambur, O.H., Frye, S.A., and Tønjum, T. (2007). New functional identity for the DNA uptake sequence in transformation and its presence in transcriptional terminators. J. Bacteriol. 189, 2077–2085. Atrouni, A. Al, Hamze, M., and Kempf, M. (2016). Reservoirs of non-baumannii Acinetobacter species. Front. Microbiol. 7, 49. Babenko, V.N., Chadaeva, I. V, and Orlov, Y.L. (2017). Genomic landscape of CpG rich elements in human. BMC Evol. Biol. 17, 19. Bae, J., Oh, E., and Jeon, B. (2014). Enhanced transmission of antibiotic resistance in Campylobacter jejuni biofilms by natural transformation. Antimicrob. Agents Chemother. 58, 7573–7575. Bailey, T.L., and Elkan, C. (1994). Fitting a mixture model by expectation maximization to discover motifs in biopolymers. Proc. 2nd Int. Conf. Intell. Syst. Mol. Biol. 2, 28–36. Baltrus, D. (2013). Exploring the costs of horizontal gene transfer. Trends Ecol. Evol. 28, 489–495. Barany, F., Kahn, M.E., and Smith, H.O. (1983). Directional transport and integration of donor DNA in Haemophilus influenzae transformation. Proc Natl Acad Sci U S A 80, 7274–7278. Barouki, R., and Smith, H.O. (1985). Reexamination of phenotypic defects in rec-1 and rec-2 mutants of Haemophilus influenzae Rd. J. Bacteriol. 163, 629–634. 92 Beauchamp, J.M., Leveque, R.M., Dawid, S., and DiRita, V.J. (2017). Methylation-dependent DNA discrimination in natural transformation of Campylobacter jejuni. Proc. Natl. Acad. Sci. 30, E8053–E8061. Bernstein, H., Byerly, H.C., Hopf, F.A., and Michod, R.E. (1985). Genetic damage , mutation , and the evolution of sex. Science (80-. ). 229, 1277–1281. Berry, J.L., Cehovin, A., McDowell, M.A., Lea, S.M., and Pelicic, V. (2013). Functional analysis of the interdependence between DNA uptake sequence and its cognate ComP receptor during natural transformation in Neisseria species. PLoS Genet. 9, 1–10. Berry, J.L., Xu, Y., Ward, P.N., Lea, S.M., Matthews, S.J., and Pelicic, V. (2016). A comparative structure/function analysis of two type IV pilin DNA receptors defines a novel mode of DNA binding. Structure 24, 926–934. Bird, A.P. (1986). CpG-rich islands and the function of DNA methylation. Nature 321, 209–213. Brockman, K.L., Azzari, P.N., Taylor Branstool, M., Atack, J.M., Schulz, B.L., Jen, F.E.-C., Jennings, M.P., and Bakaletz, L.O. (2018). Epigenetic regulation alters biofilm architecture and composition in multiple clinical isolates of Nontypeable Haemophilus influenzae. MBio 9, e01682-18. Carlson, C., Pierson, L., Rosen, J., and Ingraham, J. (1983). Pseudomonas stutzeri and related species undergo natural transformation. J. Bacteriol. 153, 93–99. Cehovin, A., Simpson, P.J., McDowell, M.A., Brown, D.R., Noschese, R., Pallett, M., Brady, J., Baldwin, G.S., Lea, S.M., Matthews, S.J., et al. (2013). Specific DNA recognition mediated by a type IV pilin. Proc. Natl. Acad. Sci. 110, 3065–3070. Chen, I., and Dubnau, D. (2004). DNA uptake during bacterial transformation. Nat. Rev. Microbiol. 2, 241–249. Cherstvy, A.G. (2009). Positively charged residues in DNA-binding domains of structural proteins follow sequence-specific positions of DNA phosphate groups. J. Phys. Chem. B 113, 4242–4247. 93 Clausen, M., Jakovljevic, V., Søgaard-Andersen, L., and Maier, B. (2009). High-Force Generation Is a Conserved Property of Type IV Pilus Systems. J. Bacteriol. 191, 4633 LP – 4638. Collins, R.F., Frye, S.A., Balasingham, S., Ford, R.C., Tonjum, T., and Derrick, J.P. (2005). Interaction with type IV pili induces structural changes in the bacterial outer membrane secretin PilQ. J. Biol. Chem. 280, 18923–18930. Cox, M.M., Goodman, M.F., Kreuzer, K.N., Sherratt, D.J., Sandler, S.J., and Marians, K.J. (2000). The importance of repairing stalled replication forks. Nature 404, 37–41. Daniel, R. (2005). The metagenomics of soil. Nature 3, 470–478. Danner, D.B., Deich, R. a, Sisco, K.L., and Smith, H.O. (1980). An eleven-base-pair sequence determines the specificity of DNA uptake in Haemophilus transformation. Gene 11, 311–318. Danner, D.B., Smith, H.O., and Narang, S.A. (1982). Construction of DNA recognition sites active in Haemophilus transformation. Proc. Natl. Acad. Sci. U. S. A. 79, 2393–2397. Davidsen, T., Rødland, E.A., Lagesen, K., Seeberg, E., Rognes, T., and Tønjum, T. (2004). Biased distribution of DNA uptake sequences towards genome maintenance genes. Nucleic Acids Res. 32, 1050–1058. Deich, R.A., and Smith, H.O. (1980). Mechanism of homospecific DNA uptake in Haemophilus influenzae transformation. MGG Mol. Gen. Genet. 177, 369–374. Domingues, S., Harms, K., Fricke, W.F., Johnsen, P.J., da Silva, G.J., and Nielsen, K.M. (2012). Natural Transformation Facilitates Transfer of Transposons, Integrons and Gene Cassettes between Bacterial Species. PLoS Pathog. 8. Dougherty, T.J., Asmus, A., and Tomasz, A. (1979). Specificity of DNA uptake in genetic transformation of gonococci. Biochem. Biophys. Res. Commun. 86, 97–104. Draskovic, I., and Dubnau, D. (2005). Biogenesis of a putative channel protein, ComEC, required for DNA uptake: membrane topology, oligomerization and formation of disulphide bonds. Mol. Microbiol. 55, 1–29. 94 Drotschmann, U., Porstendo, D., Averhoff, B., Porstendörfer, D., Drotschmann, U., and Averhoff, B. (1997). A Novel Competence Gene , comP , Is Essential for Natural Transformation of Acinetobacter sp . Strain BD413. Appl. Environ. Microbiol. 63, 4150–4157. Duzdevich, D., Redding, S., and Greene, E.C. (2014). DNA dynamics and single-molecule biology. Chem. Rev. 114, 3072–3086. Ellison, C.K., Dalia, T.N., Vidal Ceballos, A., Wang, J.C.Y., Biais, N., Brun, Y. V., and Dalia, A.B. (2018). Retraction of DNA-bound type IV competence pili initiates DNA uptake during natural transformation in Vibrio cholerae. Nat. Microbiol. 3, 773–780. Findlay, W.A., and Redfield, R.J. (2009). Coevolution of DNA uptake sequences and bacterial proteomes. Genome Biol. Evol. 1, 45–55. Fitzmaurice, W.P., Benjamin, R.C., Huang, P.C., and Scocca, J.J. (1984). Characterization of recognition sites on bacteriophage HP1c1 DNA which interact with the DNA uptake system of Haemophilus influenzae Rd. Gene 31, 187–196. Fleischmann, R.D., Adams, M.D., White, O., Clayton, R.A., Kirkness, E.F., Kerlavage, A.R., Bult, C.J., Tomb, J.F., Dougherty, B.A., Merrick, J.M., et al. (1995). Whole-genome random sequencing and assembly of Haemophilus influenzae Rd. Science (80-. ). 269, 496–512. Frye, S.A., Nilsen, M., Tønjum, T., and Ambur, O.H. (2013). Dialects of the DNA uptake sequence in Neisseriaceae. PLOS Genet. 9, e1003458. Giltner, C.L., Nguyen, Y., and Burrows, L.L. (2012). Type IV Pilin Proteins : Versatile Molecular Modules. 76, 740–772. Gonzalez-Torres, P., Rodriguez-Mateos, F., Anton, J., and Gabaldon, T. (2019). Impact of Homologous Recombination on the Evolution of Prokaryotic Core Genomes. MBio 10, e02494-18. Goodgal, S.H. (1982). DNA uptake in Haemophilus transformation. Annu. Rev. Genet. 16, 169–192. 95 Goodman, S.D., and Scocca, J.J. (1988). Identification and arrangement of the DNA sequence recognized in specific transformation of Neisseria gonorrhoeae. Proc. Natl. Acad. Sci. 85, 6982–6986. Goodman, S.D., and Scocca, J.J. (1991). Factors influencing the specific interaction of Neisseria gonorrhoeae with transforming DNA. J. Bacteriol. 173, 5921–5923. Gordân, R., Shen, N., Dror, I., Zhou, T., Horton, J., Rohs, R., and Bulyk, M.L. (2013). Genomic regions flanking E-Box binding sites influence DNA binding specificity of bHLH transcription factors through DNA shape. Cell Rep. 3, 1093–1104. Halford, S.E., and Marko, J.F. (2004). How do site-specific DNA-binding proteins find their targets? Nucleic Acids Res. 32, 3040–3052. Harrison, A., Dyer, D.W., Gillaspy, A., Ray, W.C., Mungur, R., Carson, M.B., Zhong, H., Gipson, J., Gipson, M., Johnson, L.S., et al. (2005). Genomic sequence of an otitis media isolate of nontypeable Haemophilus influenzae: Comparative study with H. influenzae serotype d, strain KW20. J. Bacteriol. 187, 4627–4636. Harteis, S., and Schneider, S. (2014). Making the bend: DNA tertiary structure and protein-DNA interactions. Int. J. Mol. Sci. 15, 12335–12363. Hepp, C., and Maier, B. (2016). Kinetics of DNA uptake during transformation provide evidence for a translocation ratchet mechanism. Proc. Natl. Acad. Sci. U. S. A. 113, 12467–12472. Hogg, J.S., Hu, F.Z., Janto, B., Boissy, R., Hayes, J., Keefe, R., Post, J.C., and Ehrlich, G.D. (2007). Characterization and modeling of the Haemophilus influenzae core and supragenomes based on the complete genomic sequences of Rd and 12 clinical nontypeable strains. Genome Biol. 8, R103. Holbrook, J.A., Tsodikov, O. V, Saecker, R.M., and Jr, M.T.R. (2001). Specific and Non-specific Interactions of Integration Host Factor with DNA : Thermodynamic Evidence for Disruption of Multiple IHF Surface Salt-bridges Coupled to DNA Binding. J. Mol. Biol. 310, 379–401. 96 Hülter, N., Sørum, V., Borch-pedersen, K., Liljegren, M.M., Utnes, A.L.G., Primicerio, R., Harms, K., and Johnsen, P.J. (2017). Costs and benefits of natural transformation in Acinetobacter baylyi. BMC Microbiol. 17, 34. Hunt, M., Silva, N. De, Otto, T.D., Parkhill, J., Keane, J.A., and Harris, S.R. (2015). Circlator : automated circularization of genome assemblies using long sequencing reads. Genome Biol. 16, 294. Illumina (2013). 16S Metagenomic Sequencing Library. Jarmer, H., Berka, R., Knudsen, S., and Saxild, H.H. (2002). Transcriptome analysis documents induced competence of Bacillus subtilis during nitrogen limiting conditions. FEMS Microbiol. Lett. 206, 197–200. Johnston, C., Martin, B., Fichant, G., Polard, P., and Claverys, J.-P. (2014). Bacterial transformation: distribution, shared mechanisms and divergent control. Nat. Rev. Microbiol. 12, 181–196. Jones, S., Heyningen, P. van, Berman, H.M., and Thornton, J.M. (1999). Protein-DNA interactions : a structural analysis. J. Mol. Biol. 287, 887–896. Kahn, M.E., Barany, F., and Smith, H.O. (1983). Transformasomes: specialized membranous structures that protect DNA during Haemophilus transformation. Proc. Natl. Acad. Sci. U. S. A. 80, 6927–6931. Kingsford, C.L., Ayanbule, K., and Salzberg, S.L. (2007). Rapid, accurate, computational discovery of Rho-independent transcription terminators illuminates their relationship to DNA uptake. Genome Biol. 8, 1–12. Kondrashov, A.S. (1982). Selection against harmful mutations in large sexual and asexual populations. Genet. Res. Cambridge Core 40, 325–332. Korotkov, K., Gonen, T., and Hol, W. (2011). Secretins: dynamic channels for protein transport across membranes. Trends Biochem. Sci. 36, 433–443. 97 Kress-Bennett, J.M., Hiller, N.L., Eutsey, R.A., Powell, E., Longwell, J., Hillman, T., Blackwell, T., Byers, B., Mell, J.C., Post, J.C., et al. (2016). Identification and characterization of msf , a novel virulence factor in Haemophilus influenzae. PLoS One 11, e0149891. Krüger, N., and Stingl, K. (2011). Two steps away from novelty – principles of bacterial DNA uptake. Mol. Microbiol. 80, 860–867. Laurenceau, R., Baconnais, S., Gault, J., Malosse, C., Dujeancourt, A., Campo, N., Chamot-Rooke, J., Cam, E. Le, Claverys, J., and Fronzes, R. (2013). A Type IV Pilus Mediates DNA Binding during Natural Transformation in Streptococcus pneumoniae. PLoS Pathog. 9, e1003473. Lethem, M., James, S.L., Marriott, C., and Burke, J.F. (1990). The origin of DNA associated with mucus glycoproteins in cystic fibrosis sputum. Eur. Respir. J. 3, 19–23. Levine, S.M., Lin, E.A., Emara, W., Kang, J., Dibenedetto, M., Ando, T., Falush, D., and Blaser, M.J. (2007). Plastic cells and populations : DNA substrate characteristics in Helicobacter pylori transformation define a flexible but conservative system for genomic variation. FASEB J. 21, 3458–3467. Li, J., Sagendorf, J.M., Chiu, T.P., Pasi, M., Perez, A., and Rohs, R. (2017). Expanding the repertoire of DNA shape features for genome-scale studies of transcription factor binding. Nucleic Acids Res. 45, 12877–12887. Li, P., Yang, C., Xie, J., Liu, N., Wang, H., Zhang, L., Wang, X., Wang, Y., Qiu, S., and Song, H. (2015). Acinetobacter calcoaceticus from a fatal case of pneumonia harboring bla NDM-1 on a widely distributed plasmid. BMC Infect. Dis. 15, 131. Ligges, U., Krey, S., Mersmann, O., and Schnackenberg, S. (2018). TuneR: Analysis of music and speech. Lorenz, M.G., and Wackernagel, W. (1994). Bacterial gene transfer by natural genetic transformation in the environment. Microbiol. Rev. 58, 563–602. Maier, B., Potter, L., So, M., Seifert, H.S., and Sheetz, M.P. (2002). Single pilus motor forces exceed 100 pN. Proc. Natl. Acad. Sci. 99, 16012–16017. Majewski, J. (2001). Sexual isolation in bacteria. FEMS Microbiol. Lett. 199, 161–169. 98 Man, W.H., De Steenhuijsen Piters, W.A.A., and Bogaert, D. (2017). The microbiota of the respiratory tract: Gatekeeper to respiratory health. Nat. Rev. Microbiol. 15, 259–270. Marçais, G., and Kingsford, C. (2011). A fast, lock-free approach for efficient parallel counting of occurrences of k-mers. Bioinformatics 27, 764–770. Mathis, L.S., and Scocca, J.J. (1982). Recognize different specificity determinants in the DNA uptake step of genetic transformation. J. Gen. Microbiol. 128, 1159–1161. Maughan, H., and Redfield, R.J. (2009). Extensive variation in natural competence in haemophilus influenzae. Evolution (N. Y). 63, 1852–1866. Maughan, H., Sinha, S., Wilson, L., and Redfield, R. (2008). Competence, DNA uptake and transformation in Pasteurellaceae. In Pasteurellaceae: Biology, Genomics, and Molecular Aspects, P. Kuhnert, and H. Christensen, eds. (Caister Academic Press), pp. 79–98. Maughan, H., Wilson, L.A., and Redfield, R.J. (2010). Bacterial DNA uptake sequences can accumulate by molecular drive alone. Genetics 186, 613–627. Meibom, K.L., Blokesch, M., Dolganov, N.A., Wu, C.-Y., and Schoolnik, G.K. (2005). Chitin Induces Natural Competence in Vibrio cholerae. Science (80-. ). 310, 1824–1827. Mell, J.C., and Redfield, R.J. (2014). Natural competence and the evolution of DNA uptake specificity. J. Bacteriol. 196, 1471–1483. Mell, J.C., Shumilina, S., Hall, I.M., and Redfield, R.J. (2011). Transformation of natural genetic variation into Haemophilus Influenzae genomes. PLoS Pathog. 7, e1002151. Mell, J.C., Hall, I.M., and Redfield, R.J. (2012). Defining the DNA uptake specificity of naturally competent Haemophilus influenzae cells. Nucleic Acids Res. 40, 8536–8549. Mell, J.C., Viadas, C., Moleres, J., Sinha, S., Fernández-Calvet, A., Porsch, E.A., St. Geme, J.W., Nislow, C., Redfield, R.J., and Garmendia, J. (2016). Transformed recombinant enrichment profiling rapidly Identifies HMW1 as an intracellular invasion locus in Haemophilus influenza. PLoS Pathog. 12, e1005576. 99 Merod, R.T. (2014). Genetic transformation of Acinetobacter baylyi in biofilms. Appl. Environ. Microbiol. 80, 7752–7757. Michod, R.E., Wojciechowski, M.F., and Hoelzer, M.A. (1988). DNA repair and the evolution of transformation in the bacterium. Genetics 118, 31–39. Mrazek, J. (2010). Comparative analysis of sequence periodicity among prokaryotic genomes points to differences in nucleoid structure and a relationship to gene expression. J. Bacteriol. 192, 3763–3772. Muller, H.J. (1932). Some genetic aspects of sex. Am. Nat. 66, 118–138. Otto, S.P., and Lenormand, T. (2002). Resolving the paradox of sex and recombination. Nat. Rev. Genet. 3, 252–261. Overballe-Petersen, S., Harms, K., Orlando, L.A.A., Moreno, J.V., Rasmussen, S., Dahl, T.W., Rosing, M.T., Poole, A.M., Sicheritz-Ponten, T., Brunak, S., et al. (2013). Bacterial natural transformation by highly fragmented and damaged DNA. Proc. Natl. Acad. Sci. U. S. A. 110, 19860–19865. Palmen, R., Vosman, B.E.N., Buijsman, P., Breek, C.K.D., and Hellingwerf, K.J. (1993). Physiological characterization of natural transformation in Acinetobacter calcoaceticus. J. Gen. Microbiol. 139, 295–305. Palmen, R., Buijsman, P., and Hellingwerf, K.J. (1994). Physiological regulation of competence induction for natural transformation in Acinetobacter calcoaceticus. Arch. Microbiol. 162, 344–352. Pamilo, P., Nei, M., and Li, W.-H. (1987). Accumulation of mutations in sexual and asexual populations. Genet. Res. Cambridge Core 49, 135–146. Pifer, M.L., and Smith, H.O. (1985). Processing of donor DNA during Haemophilus influenzae transformation: analysis using a model plasmid system. Proc. Natl. Acad. Sci. U. S. A. 82, 3731–3735. 100 Poje, G., and Redfield, R.J. (2003). Transformation of Haemophilus influenzae. In Methods in Molecular Medicine, Haemophilus Influenzae Protocols, M. Herbert, ed. (Totowa, NJ: Humana Press Inc.), pp. 57–70. Redding, S., and Greene, E.C. (2013). How do proteins locate specific targets in DNA? Chem. Phys. Lett. 570, 1–11. Redfield, R.J. (1988). Evolution of Bacterial Transformation: Is Sex With Dead Cells Ever Better Than No Sex at All? Genetics 119, 213–221. Redfield, R.J. (1993a). Genes for breakfast : The have-your-cake- and-eat-lt-too of bacterial transformation. J. Hered. 84, 400–404. Redfield, R.J. (1993b). Evolution of natural transformation: testing the DNA repair hypothesis in Bacillus subtilis and Haemophilus influenzae. Genetics 133, 755–761. Redfield, R.J., Findlay, W.A., Bossé, J., Kroll, J.S., Cameron, A.D.S., and Nash, J.H.E. (2006). Evolution of competence and DNA uptake specificity in the Pasteurellaceae. BMC Evol. Biol. 6, 1–15. Rocha, E.P.C., Cornet, E., and Michel, B. (2005). Comparative and evolutionary analysis of the bacterial homologous recombination systems. PLoS Genet. 1, 0247–0259. Rohs, R., West, S.M., Sosinsky, A., Liu, P., Mann, R.S., and Honig, B. (2009). The role of DNA shape in protein-DNA recognition. Nature 461, 1248–1253. Rohs, R., Jin, X., West, S.M., Joshi, R., Honig, B., and Mann, R.S. (2010). Origins of specificity in protein-DNA recognition. Annu. Rev. Biochem. 79, 233–269. Salzer, R., Kern, T., Joos, F., and Averhoff, B. (2016). The Thermus thermophilus comEA / comEC operon is associated with DNA binding and regulation of the DNA translocator and type IV pili. Environ. Microbiol. 18, 65–74. Sambrook, J. (2001). Molecular cloning : a laboratory manual (N.Y.: Cold Spring Harbor Laboratory Press). 101 Schwarzenlander, C., and Averhoff, B. (2006). Characterization of DNA transport in the thermophilic bacterium Thermus thermophilus HB27. FEBS J. 273, 4210–4218. Schwarzenlander, C., Haase, W., and Averhoff, B. (2009). The role of single subunits of the DNA transport machinery of Thermus thermophilus HB27 in DNA binding and transport. Environ. Microbiol. 11, 801–808. Scocca, J.J., Poland, R.L., and Zoon, K.C. (1974). Specificity in deoxyribonucleic acid uptake by transformable Haemophilus influenzae. J. Bacteriol. 118, 369–373. Seeman, N.C., Rosenberg, J.M., and Rich, A. (1976). Sequence-specific recognition of double helical nucleic acids by proteins. Proc. Natl. Acad. Sci. U. S. A. 73, 804–808. Shak, S., Capon, D.J., Hellmiss, R., Marsters, S.A., and Baker, C.L. (1990). Recombinant human DNase I reduces the viscosity of cystic fibrosis sputum. Proc. Natl. Acad. Sci. U. S. A. 87, 9188–9192. Shimizu, M., Miyake, M., Kanke, F., Matsumoto, U., and Shindo, H. (1995). Characterization of the binding of HU and IHF, hmolous histone-like proteins of Escherichia coli, to curve and uncurved DNA. Biochim. Biophys. Acta 1264, 330–336. Sinha, S., Mell, J.C., and Redfield, R.J. (2012). Seventeen sxy-dependent cyclic amp receptor protein site-regulated genes are needed for natural transformation in Haemophilus influenzae. J. Bacteriol. 194, 5245–5254. Sisco, K.L., and Smith, H.O. (1979). Sequence-specific DNA uptake in Haemophilus transformation. Proc. Natl. Acad. Sci. U. S. A. 76, 972–976. Smith, J.M. (1985). The evolution of recombination. J. Genet. 64, 159–171. Smith, H.O., Tomb, J.F., Dougherty, B.A., Fleischmann, R.D., and Venter, J.C. (1995). Frequency and distribution of DNA uptake signal sequences in the Haemophilus influenzae Rd genome. Science (80-. ). 269, 538–540. Smith, H.O., Gwinn, M.L., and Salzberg, S.L. (1999). DNA uptake signal sequences in naturally transformable bacteria. Res. Microbiol. 150, 603–616. 102 Smith, J.M., Dowson, C.G., and Spratt, B.G. (1991). Localized sex in bacteria. Nature 349, 29–31. Stella, S., Cascio, D., and Johnson, R.C. (2010). The shape of the DNA minor groove directs binding by the DNA-bending protein Fis. Genes Dev. 24, 814–826. Stone, B.J., and Kwaik, Y.A.B.U. (1999). Natural Competence for DNA Transformation by Legionella pneumophila and Its Association with Expression of Type IV Pili. J. Bacteriol. 181, 1395–1402. Straume, D., Stamsås, G.A., and Håvarstein, L.S. (2015). Natural transformation and genome evolution in Streptococcus pneumoniae. Infect. Genet. Evol. 33, 371–380. Subramani, R., Juul, S., Rotaru, A., Andersen, F.F., Gothelf, K. V, Mamdouh, W., Besenbacher, Ќ.F., Dong, M., and Knudsen, B.R. (2010). A Novel Secondary DNA Binding Site in Human Topoisomerase I Unravelled by using a 2D DNA Origami Platform. 4, 5969–5977. Suckow, G., Seitz, P., and Blokesch, M. (2011). Quorum sensing contributes to natural transformation of Vibrio cholerae in a species-specific manner. J. Bacteriol. 193, 4914–4924. Thomas, C.M., and Nielsen, K.M. (2005). Mechanisms of , and barriers to , horizontal gene transfer between bacteria. Nat. Rev. Microbiol. 3, 711–721. Treangen, T.J., Ambur, O.H., Tonjum, T., and Rocha, E.P.C. (2008). The impact of the neisserial DNA uptake sequences on genome evolution and stability. Genome Biol. 9, R60. Vos, M. (2017). Why do bacteria engage in homologous recombination ? Trends Microbiol. 17, 226–232. de Vries, J., and Wackernagel, W. (2002). Integration of foreign DNA during natural transformation of Acinetobacter sp. by homology-facilitated illegitimate recombination. Proc. Natl. Acad. Sci. 99, 2094–2099. de Vries, J., Meier, P., and Wackernagel, W. (2001). The natural transformation of the soil bacteria Pseudomonas stutzeri and Acinetobacter sp. by transgenic plant DNA strictly depends on homologous sequences in the recipient cells. FEMS Microbiol. Lett. 195, 211–215. 103 Wang, Y., and Taylor, D.E. (1990). Natural Transformation in Campylobacter Species. J. Bacteriol. 172, 949–955. Wei, J., Czapla, L., Grosner, M.A., Swigon, D., and Olson, W.K. (2014). DNA topology confers sequence specificity to nonspecific architectural proteins. Proc. Natl. Acad. Sci. 111, 16742–16747. Weinrauch, Y., Penchev, R., Dubnau, E., Smith, I., and Dubnau, D. (1990). A Bacillus subtilis regulatory gene product for genetic competence and sporulation resembles sensor protein members of the bacterial two-component signal-transduction systems. Genes Dev. 4, 860–872. Wilson, D.S. (1975). A Theory of Group Selection. Proc. Natl. Acad. Sci. 72, 143–146. Wu, D., Hugenholtz, P., Mavromatis, K., Pukall, R., Dalin, E., Ivanova, N.N., Kunin, V., Goodwin, L., Wu, M., Tindall, B.J., et al. (2009). A phylogeny-driven genomic encyclopaedia of Bacteria and Archaea. Nature 462, 1056–1060. Yella, V.R., Bhimsaria, D., Ghoshdastidar, D., Rodríguez-Martínez, J.A., Ansari, A.Z., and Bansal, M. (2018). Flexibility and structure of flanking DNA impact transcription factor affinity for its core motif. Nucleic Acids Res. 46, 11883–11897. Yin, Y., Morgunova, E., Jolma, A., Kaasinen, E., Sahu, B., Khund-sayeed, S., Das, P.K., Kivioja, T., Dave, K., Zhong, F., et al. (2017). Impact of cytosine methylation on DNA binding specificities of human transcription factors. Science (80-. ). 356, eaaj2239. Zhurkin, V., Tolstorukov, M., Xu, F., Colasanti, A., Olson, W., and Ohyama, T. (2005). Sequence-Dependent Variability of B-DNA. In DNA Conformation and Transcription, T. Ohyama, ed. (Springer Science Business Media), pp. 18–34.     104 Appendix Table	S1.		Sample	metadata		       Sample name1Sample type Biosample IDBioproject accessionDonorstrain Genome sizeFragment size range (bp)Recipient strain% of DNA recoveredUP1  (NP lg)Taken-up DNA SAMN07187224 PRJNA38759186-028NP NalR 1,914,387 1500-17000RR3117 (rec2::spec) 2.06UP2  (NP lg)Taken-up DNA SAMN07187225 PRJNA38759186-028NP NalR 1,914,387 1500-17000 RR3125 (rec2-) 1.38UP3  (NP lg)Taken-up DNA SAMN07187226 PRJNA38759186-028NP NalR 1,914,387 1500-17000 RR3125 (rec2-) 1.38UP4  (GG lg)Taken-up DNA SAMN07187227 PRJNA387591 PittGG) 1,887,046 1500-17000RR3117 (rec2::spec) 1.60UP5  (GG lg)Taken-up DNA SAMN07187228 PRJNA387591 PittGG 1,887,046 1500-17000 RR3125 (rec2-) 1.24UP6 (GG lg)Taken-up DNA SAMN07187229 PRJNA387591PittGG (RR1361) 1,887,046 1500-17000 RR3125 (rec2-) 2.48UP7 (NP sh)Taken-up DNA SAMN07187230 PRJNA38759186-028NP NalR 1,914,387 50-800RR3117 (rec2::spec) 0.64UP8 (NP sh)Taken-up DNA SAMN07187231 PRJNA38759186-028NP NalR 1,914,387 50-800 RR3125 (rec2-) 0.32UP9 (NP sh)Taken-up DNA SAMN07187232 PRJNA38759186-028NP NalR 1,914,387 50-800 RR3125 (rec2-) 0.41UP10 (GG sh)Taken-up DNA SAMN07187233 PRJNA387591 PittGG 1,887,046 50-800RR3117 (rec2::spec) 0.77UP11 (GG sh)Taken-up DNA SAMN07187234 PRJNA387591 PittGG 1,887,046 50-800 RR3125 (rec2-) 0.91UP12 (GG sh)Taken-up DNA SAMN07187235 PRJNA387591 PittGG 1,887,046 50-800 RR3125 (rec2-) 1.03UP13 (NP lg)Input DNA SAMN07187236 PRJNA38759186-028NP NalR 1,914,387 1500-17000 N/A N/AUP14 (GG lg)Input DNA SAMN07187237 PRJNA387591 PittGG 1,887,046 1500-17000 N/A N/AUP15 (NP sh)Input DNA SAMN07187238 PRJNA38759186-028NP NalR 1,914,387 50-800 N/A N/AUP16  (GG sh)Input DNA SAMN07187239 PRJNA387591 PittGG 1,887,046 50-800 N/A N/ARdrecipient DNA SAMN12049038 PRJNA387591 Rd  (recipient) 1,831,585 Not sheared Rd KW20 N/A105   1. NP lg: Long-fragment 86-028NP DNA, GG lg: Long-fragment PittGG DNA, NP-sh: Short-fragment 86-028NP DNA, GG sh: Short-fragment PittGG DNA. 							Sample name1Mean size of library fragments Un-mapped reads Mapped readsMean MAPQ scoreReads mapping only to recipient (MAPQ>0)Reads mapping only to donor (MAPQ>0)% contam-ination with Rd DNA% of reads removedMean read coverageUP1  (NP lg) 342.4 22,982 2,6468,71 47.2 122,712 2,253,377 5.16 14.9 174UP2  (NP lg) 334.0 29,276 2,386,785 47.1 151,131 1,987,921 7.07 16.7 154UP3  (NP lg) 341.3 29,111 2,962,387 47.1 226,132 2,428,674 8.52 18.0 188UP4  (GG lg) 348.8 11,134 2,183,783 46.3 145,426 1,792,255 7.51 17.9 140UP5  (GG lg) 348.2 10,423 2,149,185 46.4 366,986 1,539,930 19.25 28.3 121UP6 (GG lg) 309.5 7,732 1,227,888 46.0 97,875 978,918 9.09 20.3 77UP7 (NP sh) 227.8 203,460 5,011,237 45.8 363,683 4,006,647 8.32 20.0 303UP8 (NP sh) 219.5 211,676 3,592,281 44.7 522,918 2,591,844 16.79 27.8 197UP9 (NP sh) 247.2 201,601 7,0983,09 46.5 801,198 5,452,848 12.81 23.2 417UP10 (GG sh) 240.9 171,799 4,746,454 45.4 323,716 3,807,524 7.84 19.8 293UP11 (GG sh) 241.0 124,946 5,563,501 46.1 157,046 4,687,429 3.24 15.7 361UP12 (GG sh) 233.8 234,499 5,298,060 44.9 205,997 4,389,741 4.48 17.1 337UP13 (NP lg) 344.5 11,787 2,715,317 46.8 1,333 24,04,540 0.06 11.4 185UP14 (GG lg) 388.4 7,774 5,836,702 46.4 1,345 5,314,003 0.03 9.0 192UP15 (NP sh) 222.4 13,750 4740911 45.3 1154 3,961,509 0.03 16.4 296UP16  (GG sh) 234.0 9531 6627713 45.3 1668 4,938,510 0.03 25.5 395Rd 241.0 4564 4083313 N/A N/A N/A N/A N/A 298106 	Table	S2	uptake-prediction	position-specific	scoring	matrix	from	Mell	et	al.’s	degenerate-sequence	uptake	experiment	             			1 2 3 4 5 6 7 8 9 10A 0.071 0.094 0.423 0.632 -0.130 -0.127 -0.073 -0.064 -0.066 -0.097C -0.092 -0.035 -0.125 -0.119 -0.112 -0.122 -0.100 1.857 -0.088 -0.116G 0.058 -0.019 -0.032 -0.071 1.010 -0.127 1.607 -0.055 1.765 1.573T -0.009 -0.027 -0.085 -0.065 -0.119 0.831 -0.125 -0.054 -0.082 -0.11211 12 13 14 15 16 17 18 19 20A -0.106 -0.013 0.147 0.070 0.044 0.035 -0.021 -0.055 -0.021 0.004C -0.030 0.052 -0.066 -0.047 -0.076 -0.081 -0.036 0.013 -0.018 -0.007G -0.104 -0.061 0.028 0.002 -0.061 -0.114 -0.113 -0.080 -0.025 0.000T 0.398 0.034 -0.070 -0.016 0.129 0.268 0.262 0.165 0.071 0.00321 22 23 24 25 26 27 28 29 30 31A 0.001 -0.002 -0.013 -0.003 0.050 -0.006 0.038 -0.059 -0.062 -0.035 -0.029C 0.008 0.002 0.001 0.013 -0.060 -0.042 -0.097 -0.011 -0.028 -0.008 -0.006G 0.011 0.006 0.010 -0.004 0.063 -0.056 -0.126 -0.121 -0.076 -0.030 -0.001T -0.020 -0.006 0.002 -0.005 -0.038 0.128 0.348 0.316 0.228 0.084 0.039107  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0395493/manifest

Comment

Related Items