Overview
Our goal was to enable a controlled phage induction through a quorum sensing signal and the release of a phage from a new host organism. The whole concept is summarized in Figure 1.
Background
Why we chose to work with phages
Bacteriophages are viruses that can only infect prokaryotes, which makes them harmless to humans.1 They have co-evolved with their host bacteria for centuries and are therefore perfectly adapted to them. This is also the reason why they specifically attack only their target bacteria.2 Due to fitness costs and/or trade-offs between phage resistance and bacterial growth, evolutionary theories suggest that the likelihood of a bacteria being resistant to both antibiotics and phages is significantly reduced.3 This makes phages a promising tool in the fight against multi-drug-resistant germs.4 For these reasons, we think phages are a good way to protect a biofilm from infiltration by pathogenic bacteria.
Phages recognize their host by docking to its surface with their long tail fibers (Figure 2). This is followed by the secondary attachment through the short tail fibers, which then leads to the injection of the phage DNA (RNA) via contraction of the sheath.5 In the cell, control is then taken over by the phage, and the entire cellular machinery is switched to phage production. Phage production stops only when the cell is lysed by phage proteins and thus bursts open, releasing the newly produced phages into the environment and the lytic cycle starts all over again. In addition to the just described lytic cycle, phages can also induce a lysogenic cycle (Figure 3).6 In the lysogenic cycle, the phage DNA is incorporated into the bacterial genome and replicates with the host until it is triggered to switch into the lytic cycle. For our project, it is very important to implement a phage in our cell and only initiate its production when a pathogen is sensed. In consequence, we decided to use a lysogenic phage and exploit its lysogeny-lysis switch. In the interest to work with model organisms, we chose to use the well-studied Escherichia coli lambda phage, which is a lysogenic phage.6 In the next paragraphs we will first describe the native lambda switch and then the genetic circuit we designed to control it.
The Native Lambda Switch
After infection of Escherichia coli, the lambda phage either enters the lytic or the lysogenic cycle. Entering the lysogenic cycle, only two genes are active. The one relevant to us, called cI or lambda repressor, is expressed by the RNA polymerase binding to the PRM-promoter (Figure 4a.). To the right of cI the PR-promoter and cro are located. Cro is expressed in the opposite direction.9 Both promoters do not overlap but are connected through the tripartite right operator OR, consisting of OR1, OR2 and OR3, where OR1 and OR3 are located within these promoters (Figure 4a.). cI forms dimers that bind to OR. The first dimer binds to OR1 repressing the Cro production and enabling the cI production. The second one binds to OR2, enhancing the binding affinity of the RNA polymerase to PRM (Figure 4b.).9 When the cI concentration reaches a threshold at which OR3 is bound, its production is repressed in a self-regulating manner. Further, PR is blocked and cro cannot be transcribed. This state is almost indefinitely stable.9
The lytic cycle can be activated by causing DNA damage, evoking the cell’s SOS response. This happens through ultraviolet light or carcinogens, for example modified benzopyrene.9 Thus RecA, which normally catalyzes recombination between DNA-molecules, gets activated as a specific protease that cleaves cI monomers. The cleaved cI cannot bind to the operator and cI is no longer produced.9 To switch into the lytic cycle, the production of Cro is necessary. Cro acts as an antagonist of cI and conducts only negative regulation. It binds the same operator sites as cI but with reverse affinity, repressing the synthesis of cI. Without cI binding to the operator, the RNA polymerase can bind to the stronger promoter PR and Cro is produced (Figure 4c.). Thus, the lytic state is initiated.9
Phage Induction
Our genetic circuit for controlled phage induction
For our project, it is necessary that a functional phage is specifically expressed. Therefore, we engineered a switch that allows us to control the induction of the expression of our chosen lambda phage. Initially, we wanted to design a switch based on the production of the lambda repressor cI, leaving lambda in the lysogenic cycle.6 The next step would be the repression of cI, whereby it would no longer be produced and would therefore decay over time. The expert Prof. Dr. Grzegorz Węgrzyn drew our attention to the fact that such a switch would be too slow, as cl is a very stable protein, and its decay would take too long. Therefore, we revisited the natural switch of the lambda phage, which is triggered by the protein RecA. Normally, DNA damage is required for RecA to undergo a structural change enabling it to cleave cI, and thereby inactivating the repressor. Prof. Dr. Węgrzyn mentioned a constitutively active mutant of RecA, RecA730. This protein has a point mutation (E38K)10, which causes RecA730 to be in its active form as soon as it is expressed, thus cleaving cI independently of DNA damage. Targeted degradation of cI monomers by RecA730 should lead to a significant acceleration of our switch.
Our switch is designed to express cI located on a plasmid and herewith maintain the lambda phage in a lysogenic state (Figure 5b.). The second part of our genetic circuit is activated by recognition of quorum sensing molecules (QS molecules) of pathogens (Figure 5c.). This is accomplished by QscR (allosteric transcription factor) binding the QS molecules and the resulting complex activates the PPA1897 promoter. Through activation of this promoter the expression of recA730 and the repressor tetR, which both act as countermeasures to cI, is induced. The produced TetR binds to the PtetO promoter and herewith prevents production of cI. The produced RecA730 degrades the remaining cI, whereby the promoter of the lytic cycle in the genome of the lambda phage is no longer repressed and cro is expressed, thus leading to the activation of the lytic cycle. With the activation of the lytic cycle the expression of bacteriophages is induced. For our proof of concept, the QS-molecule induction was replaced by an IPTG induction and eGFP was used as a reporter protein for the induction of phage expression (Figure 5). For this, we placed operator sites OR1-3 of the lambda phage also containing promoter sequences upstream of the gene of eGFP.
Lab
To test our genetic switch, we first assembled the individual DNA parts together. We chose to use “Golden Gate Assembly”. This assembly method is based on type IIS restriction enzymes, that cut outside their recognition sequence. In this case, BsaI is used whereby a single-stranded overhang of 4 base pairs is created. We had to design these individual assembly sites in each of the four fragments and the backbone pDGB3_alpha_1 to perform directional ligation. The major advantage of Golden Gate Assembly is that the restriction and ligation of multiple fragments can be carried out in the same tube since the recognition sequences of the restriction enzyme are no longer present in case of a successful assembly (Figure 6).
The resulting antibiotic-resistant colonies (P1-P14) were used for inoculation of fresh medium and the plasmids were then isolated and analyzed by restriction digest to verify the successful transformation. The vector contains two EcoRI recognition sites up- and downstream of the insert. The incubated restriction digest was then analyzed using gel electrophoresis (Figure 7).
As a result, the assembled insert (ca. 3.6 kbp) should be separated from the vector (ca. 6.5 kbp) and two distinct fragments of the respective size should be visible. This is exactly what we observed with the samples from colonies P12 and P13, which were used for further testing.
These samples were sequenced to investigate if problems could occur further down the line with substitutions or frame-shift mutations in the sequence. The sequencing results showed two mutations in the 3742 bp sequence (99.95 % percentage identity). One is placed in the terminator sequence after tetR. This should not impose any problems since it only brings in an additional CG-base pair in a non-coding region without a recognition site. The second mutation on the other hand is located in the coding region of the lambda repressor cI. The amino acid Gly-54 is replaced by Asp-54. Since the new amino acid features a negatively charged side chain, this could result in a different folding behavior of one of the key enzymes in our genetic switch. However, we assume that the mutation has little or no effect on the binding affinity of cI to the PR promoter, since the following assay showed the expected results.
The phage induction module was further characterized by investigating the fluorescence output of the respective culture before and after the induction with IPTG in place of the quorum sensing molecule AHL. For this we monitored the OD600 and the fluorescence of eGFP at 485 nm and the emission at 528 nm (Figure 8).
As expected, the ratio of fluorescence per optical density does increase rapidly after the induction (195 min) with IPTG. As a result, a nearly 12-fold increase in this regard was observed, strongly indicating that our composite part does work as intended. Furthermore, sample 12 without induction showed a stable baseline over a timeframe of more than 20 h. The behavior of sample 13 seems to be the result of a low optical density, since the raw fluorescent output is not much higher than the blank background. This suggest that our system is working with little to no significant basal expression. When implemented in the lysogen MG1655, this should result in a controlled phage induction. Nevertheless, it is necessary to mention that the informative value of our analysis is limited by the number of samples. While we did test biological duplicates, our number of technical replicates amounts to five per sample per measuring point. This has been conducted two times.
We tried to decelerate the increase in fluorescence to further strengthen our assumption, that these observed effects are achieved through our construct. Therefore, we inhibited the system through the addition of anhydrotetracycline (AHT), a molecule which shows a high affinity for TetR. The resulting AHT-TetR complex is not able to repress the expression of cI (Figure 9).
There is a visible difference between the P12+IPTG sample and the P12+IPTG+AHT t1 sample, which was inhibited at the same point as the induction took place (146 min). This is true for both biological replicates. In both cases the discrepancies are greater than their respective error margins. On the other hand, when inhibited later in the measurement (214 min), the effect is less striking. While there is a difference in the P12 samples, it is not outside the error margin. With P13, the measurements are nearly identical. The fitted curve indicates, that the AHT addition after 214 min leads to an increased expression of egfp. However, this is probably the consequence of the two measurement points after 367 min and 387 min, which can be explained by measurement errors. The samples (P12-P13) inhibited at t1 achieved only 53 % of the uninhibited fluorescence output after about 50 min upon induction. In contrast, the P12-P13 mean maximum fluorescence when inhibited at t1 is about 80 % of the regular induced sample. Overall, the comparison of the induced sample and the direct inhibition indicates that TetR was bound by AHT, which resulted in the renewed expression of cI. This shifts the cI-RecA730 equilibrium and results in a decreased expression of egfp. It is worth noticing, that even when inhibited at the same time as the induction occurs, the fluorescence output is still increasing at a greater speed than the optical density. This implies, that our switch and the respective assembly could be further simplified by erasing the TetR component, if one is willing to trade about 30 min of reaction time in the system. This could serve as a starting point for further engineering using an iterative approach to optimise our genetic switch.
In conclusion, we were able to gather data which matched our expectations in nearly every sample including the controls. The only measurement which was unexpected was the little difference between the second point of inhibition compared to the induced sample without AHT. This makes us extremely optimistic about the further characterizations and applications of our composite part at a later point in time. A logical next step would be the induction of our circuit in a transformed lysogen.
Challenges
During our work on the project, we faced several challenges that had to be overcome in order to achieve the goal we set. The integration of the functioning lambda genome into Bacillus subtilis provides one of the biggest challenges. Its assembly and expression seem difficult since phages are dependent on specific host proteins and factors making the transfer of a phage into a different host a big challenge in general. B. subtilis is a Gram-positive organism and thus differs in the membrane composition from the Gram-negative E. coli11, which may have major implications on our engineering approach, e.g., on the phage release.
First, we wanted to integrate the phage genome into B. subtilis, this specifically means integration of a huge DNA fragment (approximately 50 kbp12). For this purpose, a specific integration site was necessary so that the phage genome can get replicated within the host genome.13
In order to stay close to the naturally used integration sites of B. subtilis phages, we decided to create a new integration site for our phage in this area following the advice of Dr. Anna Dragoš. Another benefit of this approach is the independence of the amyE locus which would remain free for further modification.
Secondly, we had to think about possible solutions to the problems that may arise regarding the phage expression in the new host organism. For example, the strength of the native phage promoters may be insufficient in B. subtilis and thus limit the rate of phage production. In our opinion, the most significant challenge is the adaptation of the ribosomal binding sites (RBS) from lambda phage to B. subtilis. Without a well functional RBS, ribosomes cannot initiate translation.14 Our proposed solution for this issue is shown in our outlook.
Furthermore, the decision between lytic and lysogenic behavior was crucial to control the release of phages. We wanted to ensure the modularity of our system for it to be used in a broad spectrum of possible application. Therefore, we made a repressor-antirepressor like system, that could be applied to every phage with such a system and even further as shown through our switch, which is not a repressor-antirepressor system in the typical manner. Even though we tried to keep our system modular, we cannot evidently calculate the functionality of the phage expression in a new host organism. The expression of a phage (lambda) in a new host organism (Salmonella typhimurium) has been done before, but there was some modification necessary.15 Therefore, the limited research on this field suggests, that changing the host organism of a phage is possible though it might come with unexpected problems.
Another challenge, which Dr. Anna Dragoš pointed out to us, is the possibility that phages in the lysogenic state might produce substances that are toxic for B. subtilis. That may lead to premature death of the bacterial cell without the release of functional phages. In the specific case of the lambda phage Prof. Dr. Grzegorz Węgrzyn told us that this problem should not occur because there are only two genes active in the lysogenic state and none of them are toxic.
Outlook
To overcome some of the problems and shortcomings mentioned above, we thought about the next logical steps. Certainly, one of these shortcomings is the lack of tests in Bacillus subtilis since all experiments have been established using E. coli. This helped to achieve a realistic proof of concept for the limited time we had as a result of the prevailing pandemic but came with some major shortcomings regarding our proposed implementation. One way to overcome this hurdle is to establish our phage-induction system in B. subtilis using our proposed “lambda integration site” mechanism, that will be shown in the next paragraph. Proving the assembly of functional lambda phages in a phylogenetic distant organism would be a very important next step. To achieve this, it is probably necessary to implement B. subtilis compatible regulatory regions for control of gene expression e.g., ribosome binding sites into the lambda phage genome. One tool to enable the insertion of dozens of short sequences in one step is “Multiplex Automated Genome Engineering” (MAGE).16 Using this technology and degenerate RBS oligos, the authors were able to optimize 24 genes simultaneously and to increase the production of their pathway product up to four times. This will allow for a high throughput testing to finetune the necessary delicate balance of gene expression for phage assembly. This approach could provide a first solution for the implementation of a non-native bacteriophage into a synthetically redesigned host organism such as our B. subtilis sleeper cell.
Phage Integration Site
In order to integrate the large (ca. 50 kbp) lambda genome into B. subtilis without blocking one of the standard plasmid integration sites (e.g., the amyE locus) we decided to create a new integration site in B. subtilis. This idea came up in the interview with Dr. Dragoš and with whom we remained in contact to further develop this idea. This is just one of the examples how experts shaped the development of our project. For the purpose of creating a new integration site, we wanted to clone the attB site of lambda from E. coli and then introduce it instead of the spsM gene in B. subtilis. The attB site is the genetic sequence, that is recognized by the phage and used as integration site.17 The spsM gene is naturally used as an integration site by the spß phage (B. subtilis phage), containing sporulation-promoting factors.18 By integrating the attB site into B. subtilis, the lambda’s own integrase should be able to perform the integration.17 Another factor that speaks for the success of this method is the fact that the lambda integration site works in vitro.19 Although we could not test this approach in the lab it seems like a promising way to integrate the lambda phage into the genome of B. subtilis. Testing this theory in the laboratory should easily be possible by introducing the attB site through homologous recombination into B. subtilis and then transforming the modified B. subtilis with the lambda genome. This approach could be used as a modular platform for integration of large gene fragments by adding the attP site to them.
Phage Release
Releasing the phages out of the cell is not a trivial process, because there are various cell barriers that have to be overcome. As the release is highly crucial for the success of our work, we tried to find a solution by researching how this process naturally occurs. Therefore, we decided to focus on holins, well-known proteins that contribute to the destruction of the bacterial cell envelope (BCE).20
The production of the holins takes place in the middle and last steps of the phage infection.20 The detailed steps of holin assembly are shown in (Figure 10). The holins cause lesions to the cell membrane by getting aggregated and integrated in it. This gives other proteins like lysins the opportunity to attack the murein and then destroy it.21 This way holins trigger the host cell lysis by activating the endolysins and collapsing the proton motive force.20
We focused our research on finding a proper holin that is suitable not only for E. coli but also for B. subtilis. We excluded native lambda phage holins, since holins of Gram-negativ bacteria are unable to penetrate the thick peptidoglycan layer of Gram-positive ones.20 Therefore we decided to go for the holin of the B. subtilis phage Phi29 because it has a great similarity to the holins of lambda.
The main factor, when choosing a suitable holin, was to find the protein encoding gene, which would have a high similarity with the lambda S-genes. That is where the evolutionary conserved initiation region for the holin expression comes into action. In order not to violate the system’s integrity of the lambda phage in B. subtilis we needed to find a holin with the same dual start translational motif. This implies, that the codon sequence Met-Lys-(X)-Met is located at the beginning of both necessary genes, that serve as translational starts for polypeptides with the opposite functions. The shorter polypeptide is the active holin, or lysis-effector, whereas the longer one acts as an inhibitor of holin function. All features of the dual start motif were identified in holin gene 14 of B. subtilis phage Phi29.23,24 This would allow a simultaneous expression of lambda’s holins and the holins of Phi29, making the release of lambda phage from B. subtilis probable (Figure 11).
We found out, that each gene encoding the holins of the lambda phage and phage Phi29 is preceded by two consecutive ribosome-binding sequences and has two potential start codons. Moreover, the gene arrangements of the corresponding holins and the murein hydrolase functions in both phages seem to be identical.23
Furthermore, Dr. Wadim Weber discussed with us possible modifications of holins that could help us optimize a phage release. Thus, a possible approach could be implemented via promoters of different strength. In case of further development of this project, holins will provide an opportunity for a precisely controlled release of phages and subsequent lysis of the host cell.
The next logical step after the assembly and release of functional lambda phages in a non-natural host organism would be to achieve a highly modular system, expressing specific bacteriophages as a response to a variety of biofilm invading species. While phages themselves are very specific, their genetic elements evolved modular.24 Interestingly, this also seems to be true for the system controlling the native transition of lambda phages into the lytic cycle. Since our synthetic switch is based on the lambda phage regulatory unit, our design should be compatible with a variety of bacteriophages with a minimal number of adaptations. This includes among others the exchange of the cI-RecA730 repressor-antirepressor like system with the native repressor-antirepressor system of the respective phage. This makes the modular approach a matter of genetic storage capacity for the bacteriophage genome.
References
- 1. Potera C. Phage Renaissance: New Hope against Antibiotic Resistance. Environmental Health Perspectives. 2013 Feb 1. http://dx.doi.org/10.1289/ehp.121-a48. doi:10.1289/ehp.121-a48
- 2. Poullain V, Gandon S, Brockhurst MA, Buckling A, Hochberg ME. THE EVOLUTION OF SPECIFICITY IN EVOLVING AND COEVOLVING ANTAGONISTIC INTERACTIONS BETWEEN A BACTERIA AND ITS PHAGE. Evolution. 2007;62(1):1–11. http://dx.doi.org/10.1111/j.1558-5646.2007.00260.x. doi:10.1111/j.1558-5646.2007.00260.x
- 3. Torres-Barceló C, Arias-Sánchez FI, Vasse M, Ramsayer J, Kaltz O, Hochberg ME. A Window of Opportunity to Control the Bacterial Pathogen Pseudomonas aeruginosa Combining Antibiotics and Phages Bereswill S, editor. PLoS ONE. 2014 Sep 26:e106628. http://dx.doi.org/10.1371/journal.pone.0106628. doi:10.1371/journal.pone.0106628
- 4. Górski A, Międzybrodzki R, Węgrzyn G, Jończyk‐Matysiak E, Borysowski J, Weber‐Dąbrowska B. Phage therapy: Current status and perspectives. Medicinal Research Reviews. 2019 May 7:459–463. http://dx.doi.org/10.1002/med.21593. doi:10.1002/med.21593
- 5. Dennehy JJ, Abedon ST. Adsorption: Phage Acquisition of Bacteria. In: Bacteriophages. Springer International Publishing; 2021. p. 93–117. http://dx.doi.org/10.1007/978-3-319-41986-2_2. doi:10.1007/978-3-319-41986-2_2
- 6. Łoś J, Zielińska S, Krajewska A, Michalina Z, Małachowska A, Kwaśnicka K, Łoś M. Temperate Phages, Prophages, and Lysogeny. In: Bacteriophages. Springer International Publishing; 2021. p. 119–150. http://dx.doi.org/10.1007/978-3-319-41986-2_3. doi:10.1007/978-3-319-41986-2_3
- 7. Peltomaa R, López-Perolio I, Benito-Peña E, Barderas R, Moreno-Bondi MC. Application of bacteriophages in sensor development. Analytical and Bioanalytical Chemistry. 2015:1805–1828. http://dx.doi.org/10.1007/s00216-015-9087-2. doi:10.1007/s00216-015-9087-2
- 8. Markl J, Sadava D, Hillis DM, Heller HC, Hacker SD. Regulation der Genexpression. Purves Biologie. 2019:481–510. http://dx.doi.org/10.1007/978-3-662-58172-8_16. doi:10.1007/978-3-662-58172-8_16
- 9. Ptashne M. Chapter One: The master elements of control. In: A Genetic Switch: Phage Lambda Revisited. 3rd ed. Cold Spring Harbor (NY): Cold Spring Harbor Laboratory Press; 2004. p. 160.
- 10. Vlašić I, Šimatović A, Brčić-Kostić K. Genetic Requirements for High Constitutive SOS Expression in recA730 Mutants of Escherichia coli. Journal of Bacteriology. 2011 Sep 15:4643–4651. http://dx.doi.org/10.1128/JB.00368-11. doi:10.1128/jb.00368-11
- 11. Silhavy TJ, Kahne D, Walker S. The Bacterial Cell Envelope. Cold Spring Harbor Perspectives in Biology. 2010:a000414–a000414. http://dx.doi.org/10.1101/cshperspect.a000414. doi:10.1101/cshperspect.a000414
- 12. Casjens SR, Hendrix RW. Bacteriophage lambda: Early pioneer and still relevant. Virology. 2015:310–330. http://dx.doi.org/10.1016/j.virol.2015.02.010. doi:10.1016/j.virol.2015.02.010
- 13. Tal A, Arbel-Goren R, Costantino N, Court DL, Stavans J. Location of the unique integration site on an Escherichia coli chromosome by bacteriophage lambda DNA in vivo. Proceedings of the National Academy of Sciences. 2014:7308–7312. http://dx.doi.org/10.1073/pnas.1324066111. doi:10.1073/pnas.1324066111
- 14. Breitling R, Sorokin AV, Behnke D. Temperature-inducible gene expression in Bacillus subtilis mediated by the c1857-encoded repressor of bacteriophage lambda. Gene. 1990:35–40. http://dx.doi.org/10.1016/0378-1119(90)90132-B. doi:10.1016/0378-1119(90)90132-b
- 15. Harkki A, Palva ET. Application of phage lambda technology to Salmonella typhimurium. Molecular and General Genetics MGG. 1984:256–259. http://dx.doi.org/10.1007/BF00332756. doi:10.1007/bf00332756
- 16. Wang HH, Isaacs FJ, Carr PA, Sun ZZ, Xu G, Forest CR, Church GM. Programming cells by multiplex genome engineering and accelerated evolution. Nature. 2009:894–898. http://dx.doi.org/10.1038/nature08187. doi:10.1038/nature08187
- 17. Tal A, Arbel-Goren R, Costantino N, Court DL, Stavans J. Location of the unique integration site on an Escherichia coli chromosome by bacteriophage lambda DNA in vivo. Proceedings of the National Academy of Sciences. 2014 May 5:7308–7312. http://dx.doi.org/10.1073/pnas.1324066111. doi:10.1073/pnas.1324066111
- 18. Dragoš A, Priyadarshini B, Hasan Z, Strube ML, Kempen PJ, Maróti G, Kaspar C, Bose B, Burton BM, Bischofs IB, et al. Pervasive prophage recombination occurs during evolution of spore-forming Bacilli. The ISME Journal. 2020 Dec 20:1344–1358. http://dx.doi.org/10.1038/s41396-020-00854-1. doi:10.1038/s41396-020-00854-1
- 19. Nash HA. Integrative recombination of bacteriophage lambda DNA in vitro. Proceedings of the National Academy of Sciences. 1975 Mar 1:1072–1076. http://dx.doi.org/10.1073/pnas.72.3.1072. doi:10.1073/pnas.72.3.1072
- 20. Fernandes S, São-José C. Enzymes and Mechanisms Employed by Tailed Bacteriophages to Breach the Bacterial Cell Barriers. Viruses. 2018:396. http://dx.doi.org/10.3390/v10080396. doi:10.3390/v10080396
- 21. Young R. Phage lysis: do we have the hole story yet? Current Opinion in Microbiology. 2013:790–797. http://dx.doi.org/10.1016/j.mib.2013.08.008. doi:10.1016/j.mib.2013.08.008
- 22. Dennehy JJ, Abedon ST. Phage Infection and Lysis. Bacteriophages. 2021:341–383. http://dx.doi.org/10.1007/978-3-319-41986-2_53. doi:10.1007/978-3-319-41986-2_53
- 23. Bläsi U, Young R. Two beginnings for a single purpose: the dual-start holins in the regulation of phage lysis. Molecular Microbiology. 1996:675–682. http://dx.doi.org/10.1046/j.1365-2958.1996.331395.x. doi:10.1046/j.1365-2958.1996.331395.x
- 24. Botstein D. A THEORY OF MODULAR EVOLUTION FOR BACTERIOPHAGES. Annals of the New York Academy of Sciences. 1980:484–491. http://dx.doi.org/10.1111/j.1749-6632.1980.tb27987.x. doi:10.1111/j.1749-6632.1980.tb27987.x