Evaluation of Artificial Intelligence-Generated DNA Aptamers Against Treponema pallidum Surface Proteins
JohnG.Bruno1✉Emailjbruno@nanohmics.com
ShamsudinNasaev2
DmitryUfaev2
JeffreyC.Sivils1
1
A
Nanohmics Inc6201 E. Oltorf Street Suite 40078741AustinTX2
A
A
Xelari Inc3The Green B Dover19901DE
John G. Brunoa,*, Shamsudin Nasaevb, Dmitry Ufaevb and Jeffrey C. Sivilsa
aNanohmics Inc.
6201 E. Oltorf Street
Suite 400
Austin, TX 78741
bXelari Inc.
8 The Green B
Dover, DE 19901
*Corresponding author
Email: jbruno@nanohmics.com
Abstract
DNA aptamer sequences were selected in silico with assistance from artificial intelligence (AI) using the new Xelari.com platform against three known immunogenic surface proteins of Treponema pallidum designated Tpp17, Tpp47 and Tp0751. The in silico aptamers were synthesized with 5’ Alexa Fluor 647 labels and shown to bind live Treponema pallidum by fluorescence microscopy and spectrofluorometry. While AI-predicted Kd values differed somewhat from measured Kd values, it is clear that the aptamers bound the T. pallidum surface which is consistent with the molecular docking models for each aptamer with each cognate target protein all of which are thought to be largely accessible for binding (i.e., not predominately embedded in the outer membrane or cell wall). All three aptamers also demonstrated good specificity in cross-reactivity binding studies.
Keywords:
agentic
aptamer
artificial intelligence
in silico
syphilis
Treponema
A
1. Introduction
Numerous groups have selected nucleic acid aptamer DNA or RNA sequences via in silico computational techniques [1–7]. For our in vitro therapeutic studies, we sought aptamers to bind T. pallidum and kill this pathogen which causes syphilis via opsonization [8] and complement-mediated lysis [9–11] especially because of shortages in the benzathine penicillin antibiotic supply, the lack of a syphilis vaccine and T. pallidum’s alarming resistance to penicillin alternatives such as azithromycin [12–16].
To date the only aptamers we have found in the literature that were able to bind T. pallidum were a subset of the aptamers developed against T. denticola by Park et al. [17] and these T. denticola aptamers were partially effective at opsonization of T. pallidum in our hands (DOI:10.31038/IDT.2025613). Still, we sought even higher affinity anti-T. pallidum aptamers for our novel syphilis therapeutics approaches. Thus, we chose to pursue the agentic AI-based approach now available from Xelari.com which is based on a multi-stage process for aptamer modeling using rule-based and stochastic algorithms for geometric transformations and molecular mechanics simulations, as well as artificial neural networks for energy and structural predictions.
As input data, the platform uses the 3D structure of the target protein, obtained either from the PDB database (DOI:10.1093/nar/28.1.235) or predicted using tools such as AlphaFold 2 (DOI:10.1038/s41586-021-03819-2). During processing, the platform analyzes the protein surface model to identify the most favorable binding regions. It then performs de novo design of seed structures through stochastic sampling of poses from a predefined list of nucleotides (standard deoxyribonucleotides were used in this study). For each sampled pose, a scoring function is applied to approximate the contribution of each individual nucleotide to the overall interaction energy. The greater the contribution, the more favorable the position of that nucleotide is considered to be. After the seed structures are generated, optimal frameworks are constructed by combining nucleotides with high energetic contributions into continuous chains. Following this assembly step, molecular dynamics simulations are carried out to relax the complex, and the dissociation constant (Kd) of each complex is estimated. This process yields a set of aptamers, from which the sequence with the Kd value closest to the desired target is selected.
As our initial set of protein targets, we chose three well-characterized immunogenic proteins thought to be loosely associated with the outer membrane or cell wall of T. pallidum. In particular, these proteins as shown in Table 1 are designated Tpp17, Tpp47 and Tp0751 [18–23]. The reader is advised that these proteins go by other similar names or designations in the literature, but the UniProt numbers listed in Table 1 provide clarity for each protein’s identity. The 3D structural PDB files associated with the Table 1 UniProt numbers were fed into Xelari’s platform to generate the three aptamer DNA sequences shown in Table 1 in the Results section.
A
Finally, the AI-generated aptamers were tested for binding to real live T. pallidum cells as well as related and unrelated bacterial species. As shown in this work, the three AI-generated aptamers demonstrated relatively high affinity and good specificity for binding live T. pallidum cells.2. Materials and methods
2.1. In Silico aptamer (Xelamer) AI selection criteria
PDB files for the three UniProt database target protein accession numbers given in Table 1 were loaded into separate 24 hour runs on the Xelari.com platform at https://platform.xelari.com/. Selection parameters were set as follows: Na+ concentration was 137 mM, K+ concentration was 2.7 mM, Mg2+ was 0 mM, pH was 7.2 and temperature was set to 37˚C since the aptamers may eventually be used therapeutically in vivo and need to bind well at body temperature. Target aptamer length was set to 39 nucleotides and desired binding affinity was set to the highest setting Kd = 1 nM with associated DNA parameters. The Xelari platform output DNA aptamer sequences between 37 to 42 bases with associated predicted Kd values and 3D docked aptamer-target protein PDB structures that were viewed, rotated and analyzed using RasMol version 2.7.5.2 software.
2.2. Bacteria and culture conditions
Treponema pallidum was purchased from American Type Culture Collection (ATCC; Manassas, VA) as Treponema phagedenis (ex Brumpt) Smibert, Kazan 8 strain (ATCC No. 27087) in lyophilized form. Treponema denticola (ex Flügge; ATCC No. 35404) was shipped frozen from ATCC. The treponeme cultures were rehydrated or thawed in 10 ml of Oral Treponema Enrichment Broth (OTEB) tubes from Anaerobe Systems Inc. (Morgan Hill, CA; Cat. No. AS-603) at 37°C in a slowly rotating closed incubator with a microaerophilic environment induced by Becton Dickinson Gas Pak™ pouches (Cat. No. 260680). T. pallidum cultures were capped and passaged weekly by 1:10 dilution of the previous passage in OTEB. Other bacteria were cultured from Culti-Loops™ purchased from Thermo Fisher and grown on blood agar plates overnight at 37˚C.
2.3. Dye-labeled aptamer synthesis
Each of the aptamer DNA sequences shown in Table 1 was synthesized at Integrated DNA Technologies (Coralville, IA) with a 5’ Alexa Fluor (AF) 647 dye label and purified by HPLC. Dry aptamers were rehydrated at a 100 µM concentration in sterile Dulbecco’s phosphate buffer saline (PBS) without Ca2+ or Mg2+ at pH 7.2.
2.4. Spectrofluorometry and fluorescence microscopy
Aptamer binding was validated by spectrofluorometric analysis of aptamer-AF 647 staining and fluorescence microscopy as follows. The amount of each AF 647-labeled T. pallidum aptamer from Table 1 was varied from 0.1 to 4 µM and added to 100 µl of ~ 107 live T. pallidum (5 day cultures in OTEB) after pelleting at 13,000 x g for 5 min and resuspension in 1 ml of sterile PBS without Ca2+ or Mg2+ at pH 7.2 (hereby referred to simply as PBS) for 30 min at 37˚C. Samples were then centrifuged again for 5 min at 13,000 x g. The supernates were carefully siphoned out so as not to disturb the stained bacterial pellets which were then resuspended in 1 ml of fresh PBS. The 1 ml samples were then diluted to a total volume of 3 ml in PBS in polystyrene 1 cm path length cuvettes, capped, inverted several times and immediately analyzed using a Cary-Varian Eclipse™ spectrofluorometer with excitation at 650 nm and fluorescence emission scanning from 655 to 700 nm using a photomultiplier tube (PMT) setting of 700 V. Some samples were also placed in separate wells of a sterile 6 well plate and imaged using the red channel of an Invitrogen EVOS M5000 fluorescence microscope with 400X zoom setting.
For cross-reactivity studies, all bacterial species were normalized to 1.0 absorbance unit at 620 nm using a plate reader and 100 µl of each species was stained with 2 µM of each aptamer-AF 647 conjugate in 1 ml of PBS for 30 min at 37˚C followed by centrifugal pelleting, removal of the supernates and resuspension in 3 ml of PBS in plastic cuvettes followed by five 650/665 nm ex/em point readings at a PMT setting of 600 V with mean and standard deviation determinations for each group.
2.5 Kd value prediction and measurement
The Xelari.com platform derived Kd values using an evaluation function that estimates relative Kd magnitudes based on the geometric characteristics of the complex structure and the usage conditions. At the time of the experiments, a fast evaluation function was employed, which did not fit the prediction results to experimental values. This scoring function was trained on synthetic complex formation energies without conversion to experimental units, which could lead to differences of several orders of magnitude. Nevertheless, the predicted values showed good correlation with the experimentally measured Kd values and allowed reliable relative ranking of binding strength within each group of candidates.
Some samples were analyzed by simple 650/665nm ex/em fluorescence intensity analyses using the Cary-Varian Eclipse™ spectrofluorometer at a PMT setting of 700 V as a function of aptamer concentration using 100 µl of live day 5 T. pallidum cultures with the same staining protocol as previously described. These simple emission peak reads were fed into GraphPad Prism software to generate data for the measured Kd estimations in Table 1.
3. Results
Table 1 summarizes the three aptamer DNA sequences generated by using the Xelari.com platform along with the AI-predicted Kd and actual measured Kd values obtained by fluorescence peak readings at 665 nm as a function of aptamer concentration via GraphPad Prism software. The top panels of Figs. 1–3 illustrate the 3D docking models generated by the Xelari.com platform for each predicted aptamer (ribbon structure) docked with its cognate T. pallidum protein (molecular surface structure) in the most likely binding region (i.e., “epitope”) as viewed via RasMol software. The bottom panels of Figs. 1–3 use RasMol’s monomer identification and labeling feature to reveal which nucleotides of the aptamer bound which amino acids of the target protein in the putative binding sites. By studying this information in connection with what is known about how the Tpp17, Tp47 and Tp0751 proteins are oriented with respect to the T. pallidum outer membrane and cell wall and the knowledge that these proteins are generally loosely tethered to or associated with the bacterial surface [19–23], we postulated that the majority of all three proteins are accessible for interaction and binding to their respective aptamers in solution. The 3D docked model RasMol files for each of the three aptamer-protein interactions are provided as supplemental materials for interested parties wishing to explore the binding sites and orientation of the proteins with respect to the cell surface in further detail.
Target | UniProt Number | Aptamer DNA Sequence (5’ ◊ 3’) | Predicted Kd (µM) | Measured Kd (µM) |
|---|---|---|---|---|
Tpp17 | ABO37056.1 | CCACCGCGGT TATGCCCGAT GACAGTTAAG ACGGTGG | 0.089 | 1.09 |
Tpp47 | P29723 | ATCGCGGCGT CCACACCATC CGGAATGGTT TGACGCCCGA T | 3.69 | 2.567 |
Tp0751 | O83732 | ATTAGTCGTG TCGACTCGCC CGGTAGAAGC GCGGGCTCTA AT | 0.0839 | 0.91 |
Fig. 1
Top panel – AI-generated aptamer (Xelamer) ribbon structure docked with its cognate Tpp17 protein (molecular surface structure) using the Xelari.com platform. Bottom panel – Xelamer and Tpp17 backbone structures showing the nucleotide and amino acid labels used to study binding accessibility based on known orientation of the protein from the literature. This analysis also demonstrated that the AF-647 label on the 5’ end was also likely to be free to fluoresce in solution (i.e., not quenched by proximity to amino acids).
Fig. 3
Same as for Figs. 1 and 2, but for the Tp0751 system.
Physical proof of aptamer binding began with the typical representative fluorescence microscopy appearance applicable to any of the three aptamers bound to T. pallidum as shown in Fig. 4. Figure 4A shows the appearance of live T. pallidum spirochetes under phase-contrast microscopy at 400X and the Fig. 4B shows the red fluorescing appearance of any of the three AI aptamer-AF 647 conjugates bound to the bacterial surface of live T. pallidum. It is worth noting that treponemes are known to form cyst-like aggregates in culture as previously described in the literature [24] which accounts for the clumping seen in Figs. 4A and 4B.
Fig. 4
A. 400X Phase-contrast microscopy of a live day 5 T. pallidum culture. B. Representative red fluorescent appearance of all three aptamer-AF 647 conjugates after staining live T. pallidum and washing in PBS with excitation at 650 nm (400X magnification). Bacterial aggregation is common in T. pallidum cultures [24] which accounts for the visible clumps.
We next sought to characterize the affinity of the AI-generated aptamers and to study their binding by fluorometry as a function of aptamer concentration in an attempt to determine Kd values for each aptamer. Figure 5 shows concentration-dependent binding and fluorescence responses of each aptamer from 0.1, 0.2, 0.5, 1, 2 and 4 µM aptamer binding for 30 min at 37˚C in PBS along with blanks (no aptamers; red spectral traces in panels 5A – 5C).
Fig. 5
Spectroflurometric results (measured in RFUs; relative fluorescence units) from staining of live T. pallidum with 0.1 to 4 µM concentrations of each of the three aptamer-AF 647 conjugates. Red traces correspond to blanks with no aptamer-AF 647 addition. Excitation was at 650 nm with 5 nm slits and PMT setting of 700 V.
Based on these data, we took 5 point readings of each cuvette with ex/em at 650 and 665 nm to obtain mean fluorescence emission values at 665 nm and derive the Kd values reported in Table 1 for each aptamer using GraphPad Prism software and to plot the aptamer concentration-dependent binding curves shown in Fig. 6. The measured Kd values do not correlate perfectly with the predicted Kd values derived by Xelari, but it is important to point out that bacterial particles or clumps of bacteria in suspension are not comparable to truly soluble targets in solution for the purposes of determining accurate Kd values. The particulate and clumped aggregate nature of T. pallidum [24, 25] probably also accounts for some of the rough spiked emission spectra seen in Fig. 5 despite using a smoothing function provided in the Cary-Varian software.
Fig. 6
Binding curves for each of the aptamer-AF 647 conjugates generated from fluorometric measurements measuring peak heights at 665 nm after excitation at 650 nm to generate the Kd values reported in Table 1 following analysis by GraphPad Prism software.
Finally, the three AI-generated aptamers exhibited good specificity for the Treponema genus by binding well to T. pallidum and and its close relative T. denticola, but not binding well to the unrelated bacterial species seen in Fig. 7. By comparing peak heights in Fig. 7, it is apparent that the aptamers bound approximately 3–10 times better to Treponema species versus the unrelated bacterial species.
Fig. 7
Results of a fluorometric cross-reactivity study utilizing equal numbers of T. pallidum and related (T. denticola) and unrelated bacterial species stained by 2 µM aptamer-AF 647 conjugates for 30 min at 37˚C after washing in PBS. Mean fluorescence (bar heights) and standard deviations (error bars) of 5 independent readings are plotted from ex/em peak readings at 650/665 nm.
4. Discussion
The advent of AI has provided unprecedented opportunities for rapid in silico generation of novel ligands and binding molecules such as antibodies and aptamers. The new Xelari.com platform in particular has now demonstrated rapid 24 hour DNA aptamer (Xelamer) generation ability and been validated by binding live target bacteria of medical significance (T. pallidum) with relatively high affinity (Table 1) and good specificity versus unrelated and related (T. denticola) bacterial species.
An interesting question in the present work was if aptamers generated against the entire T. pallidum proteins using their 3D structural PDB files would be able to bind these proteins if part of the protein was embedded in the spirochete surface (i.e., if the binding “epitope” was not extracellular and accessible from the outside environment). The available knowledge base suggested that all three T. pallidum surface proteins should be largely accessible to aptamer binding from solution since none of the proteins appeared to be significantly inserted into the outer lipid membrane of the cell wall [18, 25]. In particular, for Tpp17 (also called TP0435), Brautigam et al. [20] developed a structural model in which Tpp17 inserts into T. pallidum’s outer membrane in such a way that it is essentially just tethered to the cell and exposes almost all of the protein to the extracellular liquid environment, although the protein can dimerize under oxidizing conditions to limit aptamer binding. For Tpp47, the 3D docking model shown in Fig. 2 shows Domain D of the protein being distal to the aptamer binding site and Domain D with the N-terminus is known to face the outer membrane of T. pallidum [22] making the suspected aptamer binding site readily available for docking with the cognate aptamer in solution. Similarly, according to Parker et al. [23] the N-terminus of Tp0751 (Pallilysin) is anchored in the T. pallidum outer membrane by a lipid moiety, making most of that surface protein accessible from extracellular solution. In summary, the empirical binding data (Figs. 4–6) correlate well with the hypothesis that the surface proteins are readily accessible for aptamer binding in solution as predicted theoretically from Figs. 1–3.
Fig. 2
Same as for Fig. 1, but for the Tpp47 system.
Our next step in future AI-generated aptamer development will be to use the new aptamers for therapeutics experiments in animals to determine if when coupled to the Fc tail of IgG they enhance phagocytosis (opsonization) and complement-mediated lysis of T. pallidum as previously proven with other aptamer systems [8, 10, 11] in vitro. And in the future, we may be able to predict in silico using AI how nucleotide substitutions with unnatural or exotic bases or chemical modifications of aptamers may impact their binding to known surface target proteins to accelerate and enhance empirical diagnostic and therapeutic experimentation in vitro and in vivo.
Declaration of competing interest
The authors declare that with the exception of J.C. Sivils, they are all shareholders in Xelari.
A
Acknowledgement
The authors thank John Spells for maintenance of all bacterial cultures.
Electronic Supplementary Material
Below is the link to the electronic supplementary material
A
Author Contribution
J.G.B. and J.C.S. performed experiments and wrote the manuscript. S.N. and D.U. generated the aptamer DNA sequences by artificial intelligence software. All authors reviewed and edited the manuscript.
Declarations
A
Competing Interests
J.G.B., S.N. and D.U. are shareholders in Xelari.com.
A
Data Availability
The three RASMol aptamer DNA and T. pallidum surface protein 3D molecular docking PDB files are provided as supplementary materials to enable public scrutiny of statements made in the manuscript.
References
1.
Ashrafuzzaman M, Tseng CY, Kapty J, Mercer JR, Tuszynski JA A computationally designed DNA aptamer template with specific binding to phosphatidylserine. Nucleic Acid Ther, 23, pp. 418 – 26, Dec 2013.
2.
Buglak AA, Samokhvalov AV (2020) Methods and Applications of In Silico Aptamer Design and Modeling, vol. 21, Nov 10
3.
Hamada M (2018) In silico approaches to RNA aptamer design, Biochimie, vol. 145, pp. 8–14, Feb
4.
Morena F, Argentati C, Tortorella I, Emiliani C, Martino S (2021) De novo ssRNA Aptamers against the SARS-CoV-2 Main Protease: In Silico Design and Molecular Dynamics Simulation. Int J Mol Sci 22:26
5.
Nosrati M, Amani J (Dec 2021) In silico screening of ssDNA aptamer against Escherichia coli O157:H7: A machine learning and the Pseudo K-tuple nucleotide composition based approach. Comput Biol Chem 95:107568
6.
Rock M, Zouganelis GD, de Andrade AFB, Drake SJ, Alexiou A, Albrakati A et al (2022) Development and validation of anti-human Alpha synuclein DNA aptamer using computer modelling techniques-an in silico study, J Integr Neurosci, vol. 21, p. 5, Jan 21
7.
Wang Q, Guo P, Cheng W, Liu Y, Zhang Y, Huo P et al (2024) An efficient and universal In silico screening strategy for acquisition of high-affinity Aptamer and its application in analytical utility, Talanta, vol. 269, p. 125535, Mar 1
8.
Bruno JG, Carrillo MP, Crowell R (Sep 15 2009) Preliminary development of DNA aptamer-Fc conjugate opsonins. J Biomed Mater Res A 90:1152–1161
9.
Fitzgerald TJ (Sep 1987) Activation of the classical and alternative pathways of complement by Treponema pallidum subsp. pallidum Treponema vincentii Infect Immun 55:2066–2073
10.
Bruno JG A review of therapeutic aptamer conjugates with emphasis on new approaches, Pharmaceuticals (Basel), vol. 6, pp. 340 – 57, Mar 19 2013
11.
Bruno JG, Carrillo MP, Phillips T (2008) In vitro antibacterial effects of antilipopolysaccharide DNA aptamer-C1qrs complexes. Folia Microbiol (Praha) 53:295–302
12.
Ramchandani MS, Cannon CA, Marra CM (Jun 2023) Syphilis: A Modern Resurgence. Infect Dis Clin North Am 37:195–222
13.
Orbe-Orihuela YC, Sánchez-Alemán M (2022) Syphilis as Re-Emerging Disease, Antibiotic Resistance, and Vulnerable Population: Global Systematic Review and Meta-Analysis, vol. 11, Dec 15
14.
Stamm LV (2015) Syphilis: antibiotic treatment and resistance, Epidemiol Infect, vol. 143, pp. 1567-74, Jun
15.
Seghers F, Taylor MM Securing the supply of benzathine penicillin: a global perspective on risks and mitigation strategies to prevent future shortages, 16, pp. 279–282, May 1 2024.
16.
Avila-Nieto C, Pedreno-Lopez N, Mitja O, Clotet B, Blanco J, Carrillo J (2023) Syphilis vaccine: challenges, controversies and opportunities, Front Immunol, vol. 14, p. 1126170
17.
Park JP, Shin HJ, Park SG, Oh HK, Choi CH, Park HJ et al (Mar 2015) Screening and development of DNA aptamers specific to several oral pathogens. J Microbiol Biotechnol 25:393–398
18.
Peeling R, Mabey D, Kamb M, Chen X-S, Radolf J, Benzaken A (10/12 2017) Syphilis Nat Reviews Disease Primers 3:17073
19.
Akins DR, Purcell BK, Mitra MM, Norgard MV, Radolf JD (Apr 1993) Lipid modification of the 17-kilodalton membrane immunogen of Treponema pallidum determines macrophage activation as well as amphiphilicity, Infect Immun. 61:1202–1210
20.
Brautigam CA, Deka RK, Liu WZ, Norgard MV (2015) Insights into the potential function and membrane organization of the TP0435 (Tp17) lipoprotein from Treponema pallidum derived from structural and biophysical analyses, Protein Sci, vol. 24, pp. 11 – 9, Jan
21.
Brautigam CA, Deka RK, Norgard MV (Apr 1 2013) Purification, crystallization and preliminary X-ray analysis of TP0435 (Tp17) from the syphilis spirochete Treponema pallidum, Acta Crystallogr Sect F Struct Biol Cryst Commun. 69:453–455
22.
Deka RK, Machius M, Norgard MV, Tomchick DR (Nov 1 2002) Crystal structure of the 47-kDa lipoprotein of Treponema pallidum reveals a novel penicillin-binding protein. J Biol Chem 277:41857–41864
23.
Parker ML, Houston S, Pětrošová H, Lithgow KV, Hof R, Wetherell C et al (Sep 2016) The Structure of Treponema pallidum Tp0751 (Pallilysin) Reveals a Non-canonical Lipocalin Fold That Mediates Adhesion to Extracellular Matrix Components and Interactions with Host Cells. 12:e1005919
24.
Arrazuria R, Caddey B, Cobo ER, Barkema HW, De Buck J (2021) Effects of different culture media on growth of Treponema spp. isolated from digital dermatitis, Anaerobe, vol. 69, p. 102345, /06/01/ 2021
25.
Liu J, Howell JK, Bradley SD, Zheng Y, Zhou ZH, Norris SJ Cellular architecture of Treponema pallidum: novel flagellum, periplasmic cone, and cell envelope as revealed by cryo electron tomography. J Mol Biol, 403, pp. 546 – 61, Nov 5 2010.