A small-molecule chemical interface for molecular programs

Abstract In vitro molecular circuits, based on DNA-programmable chemistries, can perform an increasing range of high-level functions, such as molecular level computation, image or chemical pattern recognition and pattern generation. Most reported demonstrations, however, can only accept nucleic acids as input signals. Real-world applications of these programmable chemistries critically depend on strategies to interface them with a variety of non-DNA inputs, in particular small biologically relevant chemicals. We introduce here a general strategy to interface DNA-based circuits with non-DNA signals, based on input-translating modules. These translating modules contain a DNA response part and an allosteric protein sensing part, and use a simple design that renders them fully tunable and modular. They can be repurposed to either transmit or invert the response associated with the presence of a given input. By combining these translating-modules with robust and leak-free amplification motifs, we build sensing circuits that provide a fluorescent quantitative time-response to the concentration of their small-molecule input, with good specificity and sensitivity. The programmability of the DNA layer can be leveraged to perform DNA based signal processing operations, which we demonstrate here with logical inversion, signal modulation and a classification task on two inputs. The DNA circuits are also compatible with standard biochemical conditions, and we show the one-pot detection of an enzyme through its native metabolic activity. We anticipate that this sensitive small-molecule-to-DNA conversion strategy will play a critical role in the future applications of molecular-level circuitry.


Table of contents
Most PEN DNA based molecular networks reported so far work around 40 o C to 50 o C. Here our goal is to decrease the temperature to 37 o C, in order to bring general compatibility with most known transcription factors which come from mesophilic organisms. Initially, we observed that simply using a reported bistable switch (1) (see Supplementary  Table 1), originally designed to work at 50 o C (switch50), led to loss of function when the temperature was lowered. This is shown in Supp. Fig. 1 where the switch50 assemblies containing (on not) the cognate source template is trialled over a range of temperatures. The ability of the switch to start thanks to sTα appears to decay quickly as temperature decreases. At the same time, for temperature of around 42 o C and below, a non-specific start of the circuit (i.e. start in the absence of sTα) appears and becomes dominant. We concluded that at 37 o C the specific and non-specific start times would be bound to overlap. Figure 1. Switch50 performance of a range of temperatures. The switch was tested with (circles) or without (squares) source template at various temperatures. The expected behaviour (fast switching in presence of sT and no switching in the absence of sT) is observed down to 42 o C, although the discriminative power seems to deteriorate. Below 42 o C, no discrimination was observed for Switch50.

Method
The reaction buffer used was identical to the one reported in the main text. However, the enzyme composition was different. Vent (exo-), the nicking endonucleases Nb.BsmI and Nt.BstNBI (all from NEB) were respectively used at 80, 300 and 100 U ml -1 (respectively 4%, 3% and 1% final dilutions of the commercial stock solutions). The thermophilic 5' -> 3' exonuclease ttRecJ was purified in the laboratory, stored in Diluent A (NEB)+ 0.1% Triton X-100 at a concentration of 1.53 µM and used at 22.95 nM as before. aTα was present at 50 nM, pTα at 20 nM, rTα at 50 nM and sTα at 0.5 nM (here used simply as source template).
In this experiment, Vent exo-seemed less efficient at lower temperatures. Previous PEN-DNA work has been reported with other polymerases such as Bst full length, Bst large fragment (2,3) or Bst2.0 (warm start) (4), (all from NEB). Although, Bst Large Fragment is a thermophillic polymerase, we found, in line with others (5), that it had sufficient activity at 37 o C. Klenow was also tested in preliminary experiments, and yielded similar results. Therefore, due to its compatibility with physiological temperature, lack of 3'->5' exonuclease activity and high strand displacement activity, we selected Bst large fragment DNA polymerase for use throughout the study.
We then designed a new switch using sequences with lower melting temperature, as reported in the main text. We next had to find out the correct balance between aT and pT. Insufficient pT would result in overly weak thresholding and undesired self-starting due to spurious initiation. Excess pT would create an unnecessary drop in sensitivity as stronger stimulation would be required to start the switch. We determined 6 nM of pT to be minimally sufficient (Supp Fig. 2). We chose to proceed with 7 nM of pT to increase robustness of future experiments.  We started with the design of Source Templates (sT), a hairpin structure, containing an Nt.BstNBI site followed by the B11 signal sequence. Starting from the fully elongated form, sequential nicking, B11 dissociation and polymerase extension result in constant production of B11, proportional to sT concentration. To increase circuit start efficiency, we added 2 bp at the 3'-end, to match aT but not pT. This allowed to bypass the thresholding module and increase affinity for aT. We also shortened B11 sequence on the 5'-end to compensate for 3' extension and ease dissociation from the source template. We propagated the strategy above to Protein Sensing Templates (psTs). The same hairpin structure was taken and TrpR or LacI operator site was inserted between the Nt.BstNBI site and B11 output signal. When operator site is not occupied constant signal production occurs like above. However, this mean that the output sequence is longer as it is now catenated to a part of the operator sequence. Although this makes spontaneous dissociation unlikely (post nick), this issue is alleviated by the fact that Bst polymerase has strand displacing activity. One can observe that the LacI operator site consists of 22 bp and is 3 bp longer than TrpR site. We think this could have contributed towards worse performance of LacI-based templates, by reducing the rate of output production of the corresponding PsT. This constraint was in part mitigated by adjusting concentrations of templates. Protein sensing killer Templates (pskTs) were based on corresponding psT designs with replacement of output from B11 signal to antiB11 (aB11) sequence, similar to the pT sequence. As before, the resulting output contains part of the operator site as a tail.
Supplementary Figure 3. In detail representation of the sensing modules used in this work. Here, going from left to right along a hairpin structure: polyT region and 6 bp, providing hairpin backbone; Nt.BstNBI recognition site (underlined); operator domain for TF binding (red), within which Nt.BstNBI nicking site is indicated (red triangle); functional region consisting of pT sequence or modified B11 signal sequence (shown standalone as well). The part of the sequence that was not chemically synthesised, but filled in by DNA polymerase upon reaction initiation, is shown in gray. Dashes are used for ease of interpretation. Figure 1. Small molecule sensing occurs due to unbinding of the transcription factor in response inducer addition. Sequential nicking and extension result in production of B11 "+2". B11 or B11 "+2" can bind aT and become extended. Nicking occurs. Short DNA fragments can dissociate. B11 and B11 "+2" can also bind pT. B11 "+2" is futile. B11 binds pT preferentially and is extended to contain a sequence non-complimentary to aT. B11 or B11 "+2" can also bind rT. B11 "+2" binding is futile. B11 can be extended to permanently open the rT hairpin. All "signal" strands can be degraded by ttRecJ 5' exonuclease whilst "toolbox" strands are protected due to 5' PTO modification.  Figure  1G. We illustrate here the procedure used to determine the concentration of sT that we used throughout the work. For this we carried out a serial dilution of sT, with samples containing or not 5 nM of pskTLacI. We were expecting to see a point where sT concentration would be sufficient to activate the switch, yet would be repressible. We found that a concentration of 8 pM -the minimal attempted -activated the switch with a small delay compared to higher concentrations, for which the triggering effect tended to saturate. In addition, lower concentrations of the sT correlated to better repression by pskT. Therefore, we have chosen the lowest concentration tested. We did not go lower as in potential future emulsion work lower concentrations can be problematic.

Supplementary
In this work, we did not investigate systematically the impact of TF concentration on the circuit response. However, we expect it to affect the apparent Kd. In Supp. Fig. 9, we compare the response to L-trp for two TrpR concentrations, in the case of the inverted design with pskT (positive response). The 2-fold dilution notably affects apparent Kd which increases, as expected, but also the response becomes sharper. A further dilution to 42 nM resulted in loss of discrimination of the circuit as pskT repression became insufficient to overcome inhibition for all L-trp concentrations (not shown). This tuning opens up the possibility to adjust the sensing range and dynamics, depending on the application. By adjusting the experimental conditions, it is possible to obtain various shapes for the response of a system to its two inputs. For example, compared to the system shown in Figure  3D, reproduced here as Supp. Fig. 10A, increasing pT concentration from 1nM to 4nM, setting a higher threshold, as well as decreasing psTTrpR from 62.5 pM to 12.5pM, sets a much sharper cutoff in both dimensions (Supp. Fig. 10B).
In Supp. Fig 10C, D conditions were similar to Supp. Fig 10B, and only sensing module concentration was varied: psTTrpR=125 pM, pskTLacI=7.5 nM and psTTrpR=62.5 pM, pskTLacI=10 nM respectively. The resulting slow system in D becomes almost insensitive to IPTG input, whatever the L-trp concentration. the system in C applies a larger weight to L-trp, and a smaller to IPTG, compared to the original design.  Fig. 11A).
The system is very sensitive as doubling sT concentration (from 8 to 16 pM) has a large impact on the classification limits (Supp. Fig. 11B). To obtain the smooth log-linear summing, we used 8 pM sT but rebalanced pskTTrpR, pskTLacI, from 5 nM of each to 7.5 nM of pskTLacI and 2.5 nM of pskTTrpR. This corrects the bias towards stronger TrpR-based element and brings more samples into the window of observation as the total amount of pskT leads to less repression (taking into account that LacI-based elements are less efficient) (Supp. Fig. 11C). Finally, we made a fine adjustment to pskTTrpR=1.5nM, pskTLacI=7.5nM, to further rebalance and speed up the switch to produce the final plot reported. The positive/positive design shown in Figure 3A mixes psT and pskT. Supp. Fig. 12A shows a system starting too fast for input discrimination to occur. Decreasing the amount of psTLacI ten-fold from 2.5 nM slows down the system, but makes it little sensitive to IPTG (Supp. Fig.  12B). Another option is to enter the dynamic range is to increase pT (from 4 nM to 7 nM in Supp. Fig. 12C). This provides a reasonable window of observation, which we narrowed down to arrive to the final plot reported (Supp. Fig. 12D).

Supplementary Note 5. Enzyme activity detection and gene amplification response.
In the section we wish to have the output of the PEN-DNA circuit, instead of producing a fluorescent signal, to enable the amplification of a full-length protein-encoding gene. The goal is to obtain an enzymatic activity-to-genetic amplification circuit that can be used to build autonomous selection networks for micro-compartmentalized directed evolution protocols. We thus replace (or complement) the reporter template by two primer-producing templates (ppT), which will use the signal B11 as input, and will produce two primers targeting a given gene. After isothermal incubation, primers are then produced according to the timing of ONswitching of the PEN circuit (with earlier switching leading to more primer accumulation), and can be used during a subsequent PCR. The primer-producing templates are designed as follows: 3' to 5', it contains the B11 binding site, a Nt.BstNBI recognition/nicking site with a 4 base pair spacer, and then the primer encoding sequence (Supp. Fig. 10A).
Following switch activation, B11 signal binds on ppTs and is extended. Cycle of nicking, product dissociation and polymerase extention will ensue, like previously described for psT. The result is a linear production and accumulation of primers, conditional to switch activation (Supp. We tested a pair of ppTs (coding for a primer pair) to couple TrpB activity to gene amplification by PCR. After optimizing the conditions to enable enzymatic activity, isothermal circuit, and PCR in a single buffer, the experiment was conducted one pot, with ppTs at 10 nM and the target gene at 62 pM (see below for detailed reaction conditions). To be able to monitor the reaction in real time, we used both rT and ppT downstream of the circuit. We incubated the MP with a range of EcB concentrations at 37 o C. After full switch activation was observed for the fastest sample, we waited an additional 40 minutes (to enable primer production and switched the machine to PCR-cycling program. The PCR reaction was followed in real time thanks to an intercalating dye. We indeed observed a correlation between switching time and real-time PCR amplification rate (Supp. Fig. 11A-B). We also checked by agarose gel electrophoresis that the PCR reaction produced the expected amplification product (Supp. Fig.11C). Overall, this proves that we chemically link a catalytic activity, orthogonal to DNA, to gene amplification.