Fold preserved in the cellular environment

In-cell NMR measurements of a short dsDNA duplex (dsA2) in HeLa cells showed that all base pairs were maintained in the cell, making it a suitable candidate for selT1 studies comparing in-cell to in vitro. The dsA2 (Fig. 1A) (Alvey et al. 2014), was selected for the in-cell NMR studies because of its size (12 mer), dispersed NMR signals in the imino region and well characterized structure and dynamics (assignment (Sathyamoorthy et al. 2017), Fig. 1B). It consists of a smaller two base pair long AT stretch (A2), as longer AT- regions could potentially interact with cellular milieu (Haran and Mohanty 2009; Patsialou et al. 2005). Figure 1C shows that the imino signals from the dsA2 could be detected using a conventional jump-return NMR pulse scheme (Guéron et al. 1969). Although the jump-return spectrum from in-cell spectra matches the profile from in vitro spectra (Fig. 1B) resonances from iminos are broadened, preventing resolution between T4, T22 and T18, and G20 and G6. This indicates, however, that the basic structure in cells is the same as detected in vitro.

Fig. 1

12mer dsA2 profile A: secondary structure of the dsA2 alongside relevant base pairings visible in the imino region (~ 10–15 ppm) with the imino protons detected indicated by circles. B: in vitro jump-return spectrum of the 1 mM dsA2 in NMR buffer. C: In-cell Jump-return experiment acquired with 1024 scans (23 min 29 s) of electroporated HeLa cells containing dsDNA

Cellular quantification

Quantification of the intracellular concentration of the introduced nucleic acids reveals that the majority of dsA2 inside the cell is not bound to larger cellular complexes, as the presence of NMR signal is dependent on the free mobility of molecules. Using the FAM-labelled dsA2 to quantify the total amount in the cell lysate on a d-PAGE, we can show that the NMR-detectable fraction is comparable to the total amount of introduced dsA2 in the cell through electroporation (Fig. 2). The concentration of dsA2 inside the cell can be estimated by including standard curves of FAM-labelled dsA2 (linear correlation fitted for the standard curve Fig. S3, R2 = 0.99). The highest concentration of 0.5 pmol was omitted as it lies outside the linear range of the detector, likely saturating the detector, while all the in-cell data points lie within. An estimation of the cellular volume is needed, and therefore the diameter of n = 10 cells from Fig. S2 were measured, and the cell volumes were approximated as spherical. This resulted in 17.0 ± 7.2 µm with a volume of 2.5 ± 0.5 pL. With minimal purification steps, to reduce sample loss, we quantified the cellular concentration to be 13.6 µM ± 4.0 µM, as an average over three biological replicates (Fig. 2, Fig. S3). Our in-cell spectra are similar to those of an in vitro sample of 15 µM, and integrating the imino peaks reveals only 5% larger integrated intensity in-cell. It is important to note that integrating between sample compositions with different shimming and magnetic susceptibility means that this serves only as an approximation. This would, however, indicate that most of the introduced dsA2 remains free from larger complexes, as it is visible – indicative of free tumbling (Fig. 2), albeit with reduced resolution, which can be a result of transient interactions with larger molecules, interactions with smaller molecules, structural or sample inhomogeneity and/or different shimming properties.

Fig. 2

Signals observed arise from the free dsDNA species present inside the cell. A: SOFAST-1D spectra acquired with 2048 scans (13 min) in-cell (red) overlayed on the supernatant control (orange), indicating the signal originates from inside the cell. B: For comparison the 15 µM dsA2 recorded with the same parameters in vitro. C: 20% d-PAGE of cell lysates alongside standard curve imaged with FAM fluorophore labelled dsDNA, in which 2.5% DNA was labelled with FAM, to estimate total cellular concentration of dsDNA

Cellular viability

High viability and transfection efficiency could be determined by two complementary cell viability assays and fluorescent labelling of dsA2. Flow cytometry data indicates that the dsDNA is efficiently transfected by electroporation (Fig. S4) as 83 ± 11% of all cells carry FAM-labelled dsA2 and are viable before the NMR measurements starts. Interestingly, a discrepancy is noticeable between the viability assays, Trypan Blue microscopy viability and flow cytometry, which could arise from the settings of the signal gates, or different properties of the staining dye (Table S3). Trypan Blue staining has been shown to have different biases compared to other viability stains (Mascotti et al. 2000). This is an indication of the importance of orthogonal methods for controls, especially for fields where methods are being newly established like in-cell NMR of nucleic acids.

Subcellular localisation

Although there is a visibly higher concentration of the labelled dsA2 inside the nucleus, calculations reveal that the two molecular counts are closer to equal, due to the cytosolic volume being larger. We performed confocal microscopy on our electroporated cells and could detect that the dsA2 is distributed in both cytosol and nucleus, with visibly higher concentrations in the nucleus (Fig. S2). Upon quantification, using a nuclear diameter average of 9.6 ± 2 µm, it was revealed that 18.0 ± 25.0% of the cellular volume is nuclear, in line with other findings ranging from 1–20% (Wu et al. 2022; James and Giorgio 2000; Cadart et al. 2018), alongside observations that this value can vary within populations due to cell cycle stages or environmental effects (Moore et al. 2019; Franklin et al. 2020; Kemper et al. 2007). With a concentration ratio of 65:35 (± 7%) between nucleus and cytoplasm, this reveals that the fluorescence intensity arising from the nucleus contributes 30 ± 29% of the total signal.

dsA2 outside of cells during sample preparation is only removed upon washing

By measuring the RNA concentration in the volume of interstitial fluid in pelleted cells, it is revealed that a singular washing step leads to NMR-detectable quantities of dsA2 in the surrounding medium. Washing of the cells after electroporation needs to be sufficient to remove excess dsA2 and prevent unwanted signals before the NMR sample is prepared. In order to determine this, we estimated the medium volume surrounding cells pelleted at 300 g for 3 min by resuspending 100 million cells in in 0.6 mL of L15. The resulting volume (1.07 mL) is the sum of cell volume and interstitial medium. We estimated the HeLa cell volume to be 2.5 ± 0.5 pL, by using the diameter average from microscopy images (n = 10), (Fig. S2B). From the estimated cell volume of 100 million cells (250 µL) and the precise measurement of the total volume of a cell pellet through measurement of the volume change (0.47 mL), it was concluded that ~ 46% of the cell pellet contains medium (n = 1). This means that after washing only once in 15 mL after electroporation of 100 million cells in 400 µM dsA2, followed by resuspension in 0.6 mL medium, one would expect that ~ 1.9 µM dsA2 remains in the supernatant, just by the dilution effect. To avoid this source of error we include two washing steps with 50 mL of L15 medium following electroporation, leading to a theoretical remaining concentration of 0.002 µM dsA2.

Measurement of selective T1 using selective inversion recovery

To quantify the selT1 of T4/22 and T18 in vitro, alongside all other visible peaks, we recorded an inversion recovery experiment with selective excitation pulses sampled from 0.32 to 3000 ms (Materials and Methods) (Fig. 3). Data fitting was performed using an approximation of a mono-exponential recovery fit (Farjon et al. 2009; Szulik et al. 2014; Guéron and Leroy 1995).

Fig. 3

In vitro estimation of selT1 by selective inversion recovery. A: recorded with the selective inversion recovery pulse sequence using an in vitro 1 mM dsA2 sample in NMR buffer. Each point represents a selective inversion recovery experiment using recovery delay τ and 32 scans, resulting in experiments ranging from 6 min 3 s to 8 min 3 s. The remaining signal is T18 and the fit is shown in Fig. S5. B: Exponential fit with I = A + D*exp(-τ/selT1) of T4/22 intensities. Values normalized to the estimated equilibrium intensity. The fitted selT1 (T4/T22/T18; in vitro) is 253 ± 5 ms

T18 and T4/T22 share values in the same range 224 ± 4 ms and 253 ± 5 ms. The in vitro recorded selT1 are in the range of 107 – 330 ms (Table S1), which agrees with the range of reported values (at lower temperatures for Kearns et. al.) (Farjon et al. 2009; Behling et al. 1984).

selT1 reduced in the cellular environment

By measuring selective inversion recovery measurements on dsA2 in-cell it was observed that the selT1 is reduced in the cellular environment. To measure our selective in-cell selT1, we recorded the experiment on three biological replicates (Fig. 4A), each consisting of three technical replicates, Fig. 5, corrected for sample deterioration. Human cells can exhibit markedly distinct behaviours under only slight variations in conditions, such as changes in the cellular growth medium, hence the requirement for biological triplicates arises (Moore et al. 2019; Franklin et al. 2020).

Fig. 4

Determination of selT1 in HeLa cells. A: Three biological replicates of the selective inversion recovery experiment of dsA2 in HeLa cells, recorded at 298 K. Top to bottom is first to third replica, respectively. B: Global exponential fit with shared parameter of selT1 from the three biological replicates within the first two hours of measurement, the first technical replicate (see Fig. S6 for all spectra and Fig. S7, S8 for all fits). Exponential fit with I = A + D*exp(−τ/selT1). Values normalized to the signal at equilibrium

Fig. 5

Global analysis of selT1 of first technical replicates over three biological replicates (A). Three technical replicate spectra of biological replicate 1 (other replicates in Fig. S8), individual spectra coloured in a gradient based on delay τ. The schematic time arrow above represents the start time of a spectrum over the course of the measurement. White boxes represent SOFAST-1D spectra acquired with 2048 scans (13 min) (B). Global fits of exponential recovery for selT1 (Eq. 1) of the three technical replicates. C General workflow for the selective inversion recovery experiment from sample preparation to measurement, including experimental controls

Recovery curves were fit to all imino peaks, excluding G23/G6 (12.6 ppm), for which the peak maxima observed shifted significantly between experiments, resulting in failed fits for all experiments. This was likely due to the peaks being more separated than others, but not resolved enough to identify individual peaks. The most intense imino proton signal in our in-cell spectrum of dsA2 arises from the triple T4/22/18 peak at 13.4 ppm. Our measurements represent an average of the three.

At the low signal-to-noise ratios observed in-cell, we controlled for potential biases in each step of analysis: processing, extraction, and fitting (Kemper et al. 2007). We processed the data in three different ways: simple line broadening, FID truncation, and FID truncation with linear prediction (Fig. S9). It was observed that truncating the FID maintained sufficient resolution and led to more consistent peak maxima in chemical shifts by reducing noise, which indicates that the acquisition time was longer than necessary (Fig. S9), however, sufficient acquisition time is required for quantitative assessment of selT1, to allow for signal recovery between scans. By comparing three different processing methods and arriving at the same selT1 for the lowest intensity peak T9, we gained further confidence in the data processing methods (Fig. S9). Furthermore, we chose to fix the peak maxima at a given chemical shift, as indicated by Viles et al. (Viles et al. 2001).

Fig. 6

Workflow of the gentle supernatant sample procedure: The in-cell sample (PELLET) was roughly the size of the NMR-probe coil (VP), called “active volume”. (1) Excess medium VExcess was discarded. (2) V1 was aspirated with a Pasteur pipette and the cell pellet is resuspended by flushing V1 one time on the pellet and then repeatedly inverting the sample tube until the sample is homogeneous (3.) The cell pellet was formed by hand crank centrifugation. (4.) The supernatant V1 forms the supernatant NMR sample which contains only leaked DNA from the interstitial medium and was placed in a new sample tube. The washed cells were layered with a VExcess equivalent volume of fresh L15 medium

Three series of recovery curves (technical replicates) within each biological replica were recorded (Fig. 5). This allowed for three different fit analyses. Individual technical replicate fitting, global fitting over all technical replicates of one biological replicate, and global fitting over all biological replicates. First, each technical replicate was fit individually (Fig. S8, Table S2), and renormalised by remeasuring the 0.32 ms timepoint. Technical replicates allowed for the observation of sample change over the course of a measurement. Due to increasing fit uncertainty over the course of the measurement, there is no consistent change in the selT1 observed over the total 12 h. Second, it was possible to globally fit all technical replicates from one biological replicate (Fig. 4). The selT1 value of each replicate is compensated for sample deterioration following sample degradation and cell death (from 84% viability to 69% after 4 h), by fitting linear decay values to interleaved 0.32 ms experiments. This tracked sample decay and allowed for its correction, Fig. S10. Cell death starts after electroporation as cell viability drops from 97 ± 4% to 89 ± 9% and can be expected to further decrease during the inversion recovery NMR measurement. After 4 h (one technical replicate) of measurement the viability reduces by another ~ 4%. The global fitting is superior to individual fits, confirming common physical parameters observed in each experiment, however, the inclusion of data acquired long after sample preparation, for example technical replicate 3, is suboptimal due to the decrease in viability (Table S2). The final chosen approach involves globally fitting the first technical replicates between the three biologically replicates. This has the advantage of being acquired when the cells are closest to their physiological state, increasing the data points in this time window, alongside allowing for different delay times to be used to fully characterise the recovery.

By globally fitting the first technical replicate (spectra acquired within the first four hours of measurement) of the three biological replicates, a selective T1 (T4/T22/T18; in-cell, Tech. rep. 1, global) = 84 ± 10 ms is fitted corresponding to 1/3 of its in vitro value of 253 ± 6 ms for T4/T22 and 224 ± 3 ms for T18 (Fig. 3B). This trend was observed for all experimentally fitted peaks, with values ranging from 0.14 – 0.35-fold of the original in vitro value (Table 1).

Table 1 Presented here are the globally fitted selT1 of the first technical replicate from all three biological replicates, in vitro selT1 values and the ratio in vitro / in-cell selT1 values, to illustrate the change of selT1. Individual fits of each curve are presented in Fig.S2, 3Supernatant preparation can cause artifacts

To control for cell death and sample leakage, it is important to analyse the supernatant surrounding the cells for presence of DNA, which would lead to a non- “in-cell” signal artifact. By comparing two supernatant preparation approaches, we observe that harsher treatment of cells will generate artificial supernatant dsA2 signal. Supernatant experiments were performed on each of the three biological replicates, including a fourth cell preparation utilised for our final SOFAST vs Jump-return comparison. Two approaches to prepare the supernatant NMR sample were explored. The first includes removing the medium and then cell pellet from the Shigemi tube with a Pasteur pipette, and pelleting the cells in an Eppendorf tube, followed by resuspension with a 200 µL pipette and subsequent re-pelleting. The second, and preferred method, instead involves resuspension within the Shigemi by gentle inversion (Fig. 6). Both methods accurately recreate the in-cell sample supernatant by maintaining the total active volume and consisting of the buffer surrounding the cells. In the method where the cell pellet is moved to the Eppendorf tubes and resuspension is achieved by pipetting, larger signals are observed (biological replica 2 and 3 Fig. S1), those prepared according to a more careful protocol with gentle inversion in the Shigemi tube do not (biological replica 1 and the final comparison sample discussed in last Results section).

Improving 1D-1H spectra using measured selective T1

By using the reduced selT1, it was possible to gain signal enhancements for SOFAST-1D experiments, which yielded superior signal-to-noise when compared to delay optimised jump-return spectra. Because the selT1 observed is lower than that in vitro, this allowed for shortening the recovery time of in-cell SOFAST-1D. Of the many publications concerning the topic of in-cell NMR of nucleic acids, only a few so far use the SOFAST-1D-1H approach, including Sakamoto et. al., Broft et. al. and Yamaoki et. al (Yamaoki et al. 2022; Broft et al. 2021a; Sakamoto et al. 2021). Reported 2048 scans require 29 min and 30 s of experimental time (Sakamoto et al. 2021), but acquisition delays, interscan delay, recovery delays and other pulse sequence parameters are not reported for 1D-SOFAST, only for 2D experiments (Broft et al. 2021a). The otherwise used jump-return water suppression scheme for 1D 1H experiments requires 14 min 57 s for 1024 scans in our hands, in which the recovery time is set to 1.25*non-selT1 (T4/T22 in vitro). As our selT1 values above indicate, it is possible to shorten the recovery delay significantly, when using SOFAST experiments in-cell. To estimate the recommended recovery delay for the SOFAST approach (Schanda et al. 2005), one can obtain the interscan delay (D1) recommended for the highest signal intensity by plotting the relaxation enhancement curve (Fig. 7). As a benchmark, the in vitro selT1 and its corresponding predicted recovery time maximum was compared with experimentally derived maximum (Fig. S11. The in vitro predicted recovery time gave \(_^\) ms (upper and lower limits indicated by super- and subscript, respectively) and the experimental value is \(_^\) ms.

Fig. 7

Relaxation enhancement by SOFAST approach. The curve with selT1 values indicated, fitted for the slowest recovering peak T4/T22/T18, in-cell (black) and in vitro (blue) using Eq. (1). The optimum interscan delay for the highest sensitivity corresponding to the selT1 values are denoted. The standard error is indicated by the shaded region

For the slowest recovery time measured, T4/T22/T18, the recommended recovery time for in-cell experiments results in \(_^\) ms. Figure 7 indicates that the possible signal enhancement is 45% compared to that obtained if the experiment were run with in vitro parameters. However, the recovery time is a combination of the acquisition time, disk write time and recovery delay (D1). A large component of the sequence duration is D1. Default instrument specifications limit the spectrometer D1 to a minimum value of 30 ms, which results in the smallest possible total recovery time of 138 ms in our experimental setup arising from D1 (30 ms) + acquisition time (77 ms) + disk write time (fixed at 30 ms). Through acquiring selT1 and without modifying default instrument limitations, we could optimise the relaxation delay to 30 ms which reduced total scan time. The experimental time for our setup was 2 min 37 s for 1024 scans. Using in vitro parameters this time would be 3 min 14 s. This is 1.25 times more scans per unit time with an additional expected sensitivity gain per unit time of 6% (Fig. 7).

For G11, the fastest relaxing in-cell peak, the theoretical enhancement is 82%, but without modifying default instrument limitations would be 19%, if comparing in vitro experiments chosen for maximal G11 sensitivity. If the goal is to maximise signal-to-noise for all visible peaks, the possible enhancement would be 24% if the parameters are compared to those chosen as optimal for all iminos in-cell and in vitro, which would be that of the slowest recovering peak T4/T22/T18.

These are considerations for the fast-pulsing regime, where magnetisation does not fully recover to equilibrium before the next scan begins. This means that the experiments are semi-quantitative. In order to perform quantitative experiments, full magnetisation recovery is required, typically a value of 5 × T1 is used. Here, SelT1 optimised experiments enable a more significant decrease in scan time. For T4/T22/T18 the benefit in the quantitative regime is substantial, with 22 min 27 s for the unoptimised experiment and 7 min 5 s for the optimised experiment. This corresponds to 3 times more scans per unit time with an expected sensitivity gain of 70%.

In-cell dsA2 recovery time optimised 1H-Jump-Return versus 1H-imino-SOFAST

As a final proof-of-concept the recovery-optimised SOFAST was compared to a recovery-optimised jump-return, which was measured based on in vitro parameters (Fig. 1, Fig. 7). An improvement in acquisition speed and signal-to-noise is observed when the measured selT1 in-cell is used to optimize the SOFAST experiment, in comparison with the jump-return spectra, where parameters were optimised in vitro. The two experiments were performed on the same sample, where first the in vitro optimized jump-return, and then the in-cell optimized SOFAST were performed both with the same total experimental time Fig. 8. The frequently used jump-return water suppression scheme for 1D 1H experiments requires 23 min for only 1024 scans in our hands, in which the recovery time is set to 1.3 s, which is 1.3–6 times the non-selective T1 in vitro. The optimal recovery time for 90 degree fast-pulsing regime is 1.25*T1. In this case the recovery time is 1.3*nonselT1 of T9, but 1.9 times for T4/T22, and 6-times for the fastest peak G11. The in-cell delay-optimised SOFAST experiments allowed us to record 9216 scans within the same time, resulting in a 40% higher signal-to-noise (8.14 to 11.4 S/N) for the second fastest relaxing signal (T9), and 4% for the slowest (T4/T22/T18) (Figs. 1C).

Fig. 8

A quantitative comparison of the SOFAST pulse program (red spectrum) with the Jump-Return echo sequence seen in Fig. 1 (orange spectrum) on the same sample. These 1H spectra of dsA2 in HeLa cells were recorded in 23 min 39 s (9216 scans) and 22 min 29 s (1024 scans), respectively