The combination of bPAC and the ERK-KTR reporter provides quantitative single cell ERK activity measurements that are highly sensitive to intracellular cAMP/PKA dynamics

We first wanted to test the ability of bPAC to elicit signaling in canonical cAMP pathways for which reliable single cell-reporters exist. To do so, we subjected an all-emitter monolayer (Fig 1) containing either PRKACA or ERKKTR reporters to pulses of blue light inducing bPAC in every cell. Blue light-activated bPAC was clearly able to activate PKA as measured by the PRKACA reporter [11] (Fig 2A, see Materials and methods for details). It was also able to reduce ERK pathway activity, as measured by increased Nuclear/Cytoplasmic ratio in the ERK-KTR reporter (Fig 2B). For both reporters, consecutive pulses of blue light (40 ON/40 OFF pulses) resulted in repeatable reporter responses. Furthermore, activation of bPAC increased the frequency of Ca²⁺ spikes [36] as expected (S1(A) Fig, measured using the GcAMP5h reporter [37, 38], see Materials and methods for details).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. bPAC induced emitters in all-emitter monolayers tend to have a simple, time-dependent ERK inhibition response.

(A) Periodic activation of bPAC (lower plot) results in periodic activation of PKA in single cells (upper plot). (B) Periodic activation of bPAC (lower plot) results in periodic ERK inhibition visualized by the increased ERK Kinase Translocation Reporter (ERK-KTR) nuclear to cytoplasmic ratio N/C in single cells (upper plot). (C) Time dependent snapshots of ERK-KTR reporter from experiment presented in (B). Here nuclear localization correlates with bPAC activity. (D) Periodic activation of bPAC (lower plot) in the presence of PDE inhibitor (IBMX) exhibits slower turn off of ERK inhibition (upper plot: blue, population mean) after each pulse than with no drug (upper plot, grey, normalized to blue for comparison). (E and F) Calculation of the population mean and standard deviation of the ERK-KTR N/C signal under different normalization approaches. Top 3 plots: For a given normalization approach, each single cell ERK-KTR N/C signal is shifted such that the signal at the beginning of the first pulse is zero and then normalized. Mean and standard deviation are calculated to just before the start of the second pulse. This process is the then repeated for the second pulse enforcing the standard deviation is zero at the beginning of each pulse. Top plot: mean ± standard deviation. Second plot from top: mean ± standard deviation calculated after each shifted signal is normalized by difference between the peak value of the signal during the first pulse and the value at the beginning of the first pulse. This removes the effects of peak amplitude variability in the first pulse. Third plot from top: same as prior normalization approach of the first pulse, but applied to the second pulse instead. This removes the effects of peak amplitude variability in the second pulse. Bottom plot: bPAC input pulse sequence. (E) All-emitter monolayer with identical bPAC input pulse amplitudes. (F) All-emitter monolayer with increasing bPAC input pulse amplitudes. For both (E) and (F), standard deviation is most constrained during the chosen pulse that removed the effects of peak amplitude variability (highlighted regions). Note that for some experiments one observes a decaying transient in the ERK-KTR N/C signal during the first 40 minutes when the the light is off (see single emitter examples in (B)). This occurs from exciting bPAC when searching for suitable cells to image with light that overlaps the spectrum that excites bPAC.

https://doi.org/10.1371/journal.pcbi.1009873.g002

While all three reporters measured repercussions of the activation of bPAC by blue light, the ERK-KTR was the most robust and had the larger dynamic range. In addition, it also does not interact directly with cAMP and PKA thereby preventing potential altering of endogenous cAMP/PKA dynamics. Therefore, given that we are perturbing only cAMP in this work, we used the ERK-KTR as a downstream reporter of the cAMP/PKA pathway response to bPAC induction. Specifically, we quantified the ratio of ERK-KTR nuclear concentration to the ERK-KTR cytosolic concentration (ERK-KTR N/C), which is a measurement of ERK inhibition. An increase in cAMP levels increases PKA and Epac signaling, resulting in increased ERK inhibition and diminishing ERK’s ability to keep ERK-KTR out of the nucleus [32]. As a result, the ERK-KTR nuclear concentration increases. The opposite occurs after bPAC inactivates and cAMP levels decrease (Fig 2C). Thus, the ERK-KTR N/C signal faithfully responds to bPAC activity and hence cAMP concentrations (Fig 2B).

Inhibition of phosphodiesterase (PDE) activity by IBMX leads to slower ERK-KTR N/C turn-off dynamics than cells without IBMX (Fig 2D). Therefore, the ERK-KTR N/C signal is able to mirror the slower degradation rate of cAMP ensuing from PDE inhibition [39], further evidence that this signal is sensitive and responsive to the time-dependent cAMP concentration. Different blue light input pulses further corroborate the sensitivity of the ERK-KTR N/C signal to bPAC/cAMP dynamics for the operational regime needed for our investigations (S1(B) Fig with discussion in S1 Text).

bPAC induced emitters in all-emitter monolayers have a simple, first-order time-dependent ERK inhibition response

We sought to characterize the mean and variability across the population of single cell time-dependent ERK-KTR signals in all-emitter monolayer tissues for different pulsed experiments. For all experiments, we again use a sequence of two bPAC input pulses (40 min ON/OFF). For pulses of the same amplitude, the time-dependent mean of the ERK-KTR N/C response is very similar across pulses (Fig 2E and S2(A) Fig, top plot). For successive pulses of increasing amplitude, the maximum mean ERK-KTR N/C response tracks linearly (approximately) in the pulse amplitude (Fig 2F and S2(B) Fig, top plot). In all cases, significant variability across cells exists however. Although this variability is similar to what is observed from other inhibitions on the ERK pathway [32], we sought to better understand its origins.

For the initial mean and variability (standard deviation) calculations above, we calculated the standard deviation during each bPAC input pulse separately (including during shutoff). To ensure that we are comparing only variability that is driven by bPAC over each pulse and during shutoff, we first removed baseline variability at the beginning of a given bPAC input pulse. This is achieved by shifting each single-cell ERK-KTR N/C signal so that its value at the beginning of the pulse is zero (further details in Fig 2 and S2 Text).

Next we calculated the mean and standard deviation after normalizing each single cell signal by the peak response during the first pulse (further details in Fig 2 and S2 Text). We observe that that the variability about the mean during the first pulse becomes highly constrained (Fig 2E and 2F and S2(A) and S2(B) Fig, second plot from top (highlighted region)), significantly more so than during the second pulse. Similarly, applying the same normalization approach to the second pulse instead, we observe that the variability about the mean during the second pulse becomes highly constrained, significantly more so than during the first pulse (Fig 2E and 2F and S2(A) and S2(B) Fig, third plot from top (highlighted region)). This implies that the un-normalized variability (Fig 2E and 2F and S2(A) and S2(B) Fig, top plot) is dominated by variability in the peak amplitude of the ERK-KTR N/C signal (see S2 Text for detailed analysis and discussion). And aside from peak-amplitude variability, the single cell ERK-KTR N/C signals during a pulse tend to all have a very similar time-dependent behavior.

Overall, we observe that in the uniform all-emitter monolayer, the time-dependent ERK-KTR N/C response follows a simple over-damped rise upon bPAC activation at any amplitude we tested (see S1 and S2 Movies for visual examples). This characteristic response is general across cells in the all-emitter monolayer. Crucially, these results give us a reference for comparison when assessing signal behaviors in localized groups of emitter cells surrounded by receiver cells (Fig 1, small-emitter-cluster monolayer).

Emitters in small emitter clusters can have a qualitatively different ERK inhibition signal during bPAC induction than those in an all-emitter monolayer

We next wanted to induce bPAC in small clusters of emitters surrounded by receivers to see how the emitter ERK-KTR N/C signal dynamics compared with the simple dynamics observed from emitters in an all-emitter monolayer. We generated monolayers with small emitter clusters (each emitter contains bPAC and ERK-KTR constructs) surrounded by receivers (each receiver contains an ERK-KTR construct). These mosaic monolayers were created by mixing the two strains with much smaller proportions of emitter cells. In Fig 1, we defined an ideal small emitter cluster to be a concentrated cluster of emitters surrounded by receivers. While most emitters will be touching other emitters, we also allow ‘stray’ emitters, separated from the main group of connected emitters by a receiver cell, to be part of the cluster.

Experiments containing small clusters of emitter cells surrounded by receiver cells can have portions of emitters exhibit a distinct, dynamic overshoot in their individual ERK-KTR N/C responses. Fig 3A presents a striking example of the overshoot exhibited by many of the single emitters. Specifically, after the start of each bPAC input pulse, we observed a rapid increase in the inhibition of ERK activity in the emitter cells, where activity tended to partially recover over the following few minutes despite sustained blue light illumination. (Fig 3A, single emitters (bottom panel), emitter average (top panel, top plot)). For this example, overshoot of ERK inhibition occurs in both pulses of the bPAC input (40 min. pulse width, max amplitude). Furthermore, this emitter behavior in the small emitter cluster diverges from the simple first order response observed in emitters from all-emitter monolayer experiments for the same input pulse sequence (Fig 2E, for example).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Single cell emitter ERK-KTR N/C signals from small-emitter-cluster monolayers can be qualitatively different from emitters in all-emitter monolayers.

Images of monolayers in (A) and (E): merged image of green (filtered ERK-KTR with horizontal line detector filter), red (nuclear maker), blue (nuclear bPAC marker) where emitter nuclei are purple (red+blue), and receiver nuclei are red. Labeled nuclei: white label (emitter), yellow label (receiver). (A) Results for a small-emitter-cluster monolayer. Left top panel: top plot: small-emitter-cluster average of ERK-KTR N/C signal, middle plot: adjacent receivers average of ERK-KTR N/C signal, bottom: bPAC input. Right image: small emitter cluster. Bottom panel: single cell emitter measurements of ERK-KTR N/C signal from the small-emitter-cluster experiment. (B) Time dependent snapshots of ERK-KTR reporter from experiment presented in (A) for the emitter (white ‘E’ label) whose response is plotted in the bottom single cell emitter plot in (A) (dashed vertical lines depict snapshot times) and an adjacent receiver (yellow ‘R’ label). (C) Illustrations depicting measurements used for time-to-max, time-to-ss, and overshoot-ratio statistical metrics. The time-to-max is the time to reach S_max, max signal during pulse. The time-to-ss is the time to reach S₃₀, i.e. when the signal is first within 30 percent of S_end (end-of-pulse signal value) relative to S_max. Given S_start (start-of-pulse signal value), the overshoot ratio is . Top panel: Measurements of the ERK-KTR N/C signal for a single emitter from small-emitter-cluster experiments (including that in (A)). Bottom panel: Measurements of the ERK-KTR N/C signal for a single emitter from all-emitter experiments (including those in Fig 2E and 2F). As shown in both panels, measurements are taken during each bPAC input pulse. (D) Scatter plots of time-to-max times, time-to-ss times, and overshoot-ratio for single cell ERK-KTR N/C signals taken from all-emitter experiments (blue, with the two rightmost ones from experiments presented in Fig 2E and 2F), small-emitter-cluster experiments (green, with the leftmost one from experiment presented in (A)), and combined data for 10 single-emitter-cluster experiments (some experiments have more than one single emitter cluster in the field of view). Mean (magenta diamond) ± standard deviation (magenta error bars), error bars in the mean (green, standard deviation/) representing 95 percent confidence interval in the mean, and median (yellow diamond) are also plotted. Number of data points N used are displayed below each group label. (E) Results for a single-emitter-cluster experiment. Left panel: top plot: single emitter ERK-KTR N/C signal, middle plot: adjacent receivers average of ERK-KTR N/C signal, bottom plot: bPAC input. Right image: single emitter cluster. Bottom panel: plots of ERK-KTR N/C signal for adjacent receivers, and bPAC input (bottom plot).

https://doi.org/10.1371/journal.pcbi.1009873.g003

While the small-emitter-cluster example in Fig 3A shows a majority of emitters exhibiting strong, dynamic overshoot in both pulses, other small emitter clusters may have smaller proportions of emitters respond with this type of overshoot (S3(A) Fig, left and center panels). Furthermore, some emitters may exhibit strong overshoot during one pulse, but exhibit a more flat response in the other. And some emitters may exhibit a flat response during both pulses (S3(A) Fig, left and center panels). That is, for a given small-emitter-cluster experiment, there can be heterogeneity in the ERK-KTR N/C signal across emitters, as well as across bPAC input pulses for a given emitter. It is both an emitter’s ability to exhibit overshoot as well as the observed response heterogeneity within a given small emitter cluster that makes this overall phenotype intriguing.

For the example in Fig 3A, the ERK-KTR N/C overshoot within single emitters from the small emitter cluster tends to take anywhere from 8–15 minutes to complete, i.e. there is variability (Fig 3A, single emitter responses. Also see S3 Movie, includes description). On the other hand, all-emitter experiments typically show very little ERK-KTR N/C overshoot, statistically (Fig 2E and 2F, with single emitter plots in Fig 2B). When they do, it tends to be at a much slower timescale and a more subtle, slow overshoot (S1(C) Fig, fourth single cell plot, for example) as compared to the dramatic, dynamic overshoot in many of the emitters in the small-emitter-cluster examples.

Statistical analysis of emitter ERK-KTR signals supports the qualitative differences observed between all-emitter and small-emitter-cluster experiments

Next we wanted to statistically quantify the qualitative differences observed in the ERK-KTR N/C signal between emitters in all-emitter experiments and those in small-emitter-cluster experiments. The first statistical metric we calculated for each emitter signal was the time to reach peak ERK-KTR N/C signal amplitude during the bPAC input pulse, denoted as time-to-max (see Fig 3C for illustration). This is a proxy for when overshoot occurs in emitters or when it reaches steady state when there is no overshoot. The time-to-max tends to be significantly slower on average for emitters in an all-emitter monolayer than for emitters in a small emitter cluster surrounded by receivers (Fig 3D, left panel, mean error bars in green represent 95 percent confidence interval of the mean, compare means across groups). Statistically, the all-emitter experiment with the lowest mean (bPAC input pulse amplitudes of one (max)) has a greater mean than the pooled small emitter cluster data (p < 10⁻⁵ (see Materials and methods for details)).

Next, for each emitter, we calculated the approximate steady-state convergence time after reaching the maximum signal amplitude. We defined this as the time the signal drops to within 30 percent of the end of pulse signal amplitude relative to the max signal amplitude, and denoted this time as time-to-ss (see Fig 3C for illustration). Thus, for a given emitter, time-to-ss ≥ time-to-max. For the all emitter experiments, we see very little change in the mean between the time-to-ss and the time-to-max, since the mean time-to-max is already high and cannot exceed 40 minutes (Fig 3D, middle panel, all-emitter data). This is also because for many of these emitters the time-to-max represents reaching steady-state not overshoot. For the small-emitter-cluster experiments we see the mean time-to-ss shift by less than 10 minutes on average (across experiments) relative to the time-to-max (Fig 3D, middle panel, small-emitter-cluster data, dashed line across left and middle panels serves as a reference). Because many of the emitters exhibit overshoot, this give an average sense of the overshoot dynamics, with plenty of variability about this mean (see S3(C) Fig for time-to-ss−time-to-max statistics, mean and variability).

For a final metric we wanted to quantify the magnitude of the overshoot across experiments. For this, we applied an overshoot ratio metric defined as where S_max is the max signal value during the bPAC pulse, S_start is signal value at the start of the pulse, and S_end is the signal value at the end of the pulse (see Fig 3C for illustration). For all-emitter experiments the majority of the overshoot-ratio measurements are near one, with a few outliers (Fig 3D, right panel, all-emitter data). While for emitters in small emitter clusters, there is a wide range of overshoot-ratio measurements ranging from near one, similar to the all-emitter experiments, to very large overshoot-ratio (outside the range plotted) (Fig 3D, right panel, small-emitter-cluster data, some data points are outside the range plotted). Statistically, the pooled small emitter cluster data has a greater mean for the overshoot-ratio than any all-emitter experiments (p < .012 (see Materials and methods for details)).

In summary, emitters in small-emitter-cluster experiments have a larger overshoot ratio on average than those in all-emitter experiments. And for the emitters in small-emitter-cluster experiments, the overshoot dynamics on average converge within 10 minutes after the peak signal is reached with some emitters having faster dynamics than the average and some having slower dynamics.

Isolated single emitters also can exhibit dynamic overshoot in their ERK inhibition signal during bPAC induction

Importantly, for the small-emitter-cluster experiment, the surrounding receiver cells exhibit a response in their ERK-KTR N/C signal indicating coupling between emitter and adjacent receiver cells, all of which touch emitters. And on average, the receiver cells do not exhibit overshoot (Fig 3A, top panel, middle plot). An example of overshoot in the ERK-KTR nuclear concentration is captured for a single emitter along with its coupling to an adjacent receiver which exhibits no overshoot (Fig 3B). This motivated us to induce bPAC in the most minimal small-emitter-cluster, a single emitter, i.e. single emitter cluster, where there are no direct physical interactions or contact with other emitters. We conjectured that it might also exhibit overshoot. That is, the overshoot in emitters is somehow fundamentally related to the coupling of the ERK-KTR N/C signal to receiver cells, possibly through cAMP coupling. For this paper, we examine single emitters that are isolated by at least three or more receiver cells between itself and other single emitters or small emitter clusters (Fig 3E, image: single emitter (white label), adjacent receiver cells (yellow labels)). Indeed, when we apply a two pulse bPAC input, both at max amplitude, we observe overshoot in the ERK-KTR N/C signal, most prominently during the second pulse (Fig 3E, single emitter plot. Also see S4 Movie, includes description). The first pulse yields a quick minimal overshoot, one could argue, while the second pulse yields a slower more pronounced overshoot, similar to those observed in some of the single cells in the small cluster emitter experiment (Fig 3A). While the emitter does exhibit overshoot, the response is quite heterogenous across pulses. As with small emitter clusters, there are also single emitter clusters that exhibit overshoot across pulses, as well as ones that exhibit no overshoot (S3(A) Fig, right panel, and see S3(B) Fig for images of time snapshots of emitter ERK-KTR response for emitters that exhibit overshoot).

Similar to the small-emitter-cluster experiments, pooled data from numerous single-emitter-cluster experiments show that time-to-max in the ERK-KTR N/C signal during a bPAC input pulse for single emitters tends to be lower, on average, than emitters in the all-emitter experiments (Fig 3D, left panel, single-emitter-cluster data, mean error bars in green represent 95 percent confidence interval of the mean, compare means across groups. Statistically, the all-emitter experiment with the lowest mean (bPAC input pulse amplitudes of 1 (max)) has a greater mean than the pooled single emitter cluster data (p < 10⁻⁵ (see Materials and methods for details)). In addition, both time-to-ss and overshoot ratio for the pooled single-emitter-cluster experiments behaved similarly to the small emitter cluster data (compare in Fig 3D, middle panel (time-to-ss) and right panel (overshoot ratio), and in S3(C) Fig for time-to-ss−time-to-max statistics, mean and variability). Statistically, the pooled single emitter cluster data has a greater mean for the overshoot ratio than any all-emitter experiments (p < .012 (see Materials and methods for details)).

For single-emitter-cluster experiments, unlike the emitter, the average ERK-KTR N/C signal of the adjacent receiver cells tend to not exhibit a pronounced overshoot during either pulse (Fig 3E, top panel, middle plot and S3(B) Fig, plots of average of adjacent receivers). This also tends to be the case for individual receivers (Fig 3E, lower panel: single receiver cells). For the single emitter example in Fig 3E, we also observe a rare example of an adjacent receiver (cell 37) entering prophase of mitosis losing its nuclear envelope and, thus, the ERK-KTR is cytosolic and unresponsive. Later in this work we will discuss how cells in this state receive little to no coupled ERK inhibition signal from the emitter.

Finally, we wanted to look at the at ERK-KTR N/C signal propagation from single emitter clusters as a function of steady-state amplitude taken at the end of a 40 minute pulse. We found that steady-state receiver signals follow a decreasing trend with increasing distance from the emitter (S3(D) Fig).

Coupling of ERK-KTR N/C signal requires cAMP transport through gap junctions controlled by PKA

We next wanted to explain the overshoot observed in the emitter ERK-KTR N/C signal for small emitter and single emitter clusters as well as the behavior of the ERK-KTR N/C signal propagation to receiver cells. We hypothesized that the cAMP produced by bPAC was diffusing through gap-junctions to the surrounding receiver cells which were acting as sinks, depressing maximal cAMP concentration in the emitter cells. Indeed, inhibiting gap-junctions inhibits the receiver ERK-KTR N/C signal (Fig 4A (average signals) and S4(B) Fig (mean with error bars)), implicating gap junctions as the main coupling mechanism, and thus cAMP as the coupled signaling molecule between emitters and receivers. Interestingly, the average emitter ERK-KTR N/C signal decay after the bPAC input pulse shuts off (Fig 4A) is very similar to all-emitter experiments with IBMX (compare Fig 2D with S4(A) Fig). This is likely because carbenoxolone is known to inhibit PDEs resulting in slower cAMP decay.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. PKA and cell-cell coupling of cAMP through gap junctions are required for the dynamic overshoot observed in the ERK-KTR N/C signal of bPAC induced emitters from small emitter clusters.

Single cell ERK-KTR N/C signals in all-emitter monolayers are representative of the dynamics due to intracellular regulation. (A-D) Left image depicts emitter cluster (emitter cells—purple nuclei with white labels, adjacent receiver cells—red nuclei with yellow labels). Right Panel: top plot: average small-emitter-cluster ERK-KTR N/C signal, middle plot: average adjacent receivers ERK-KTR N/C signal, bottom plot: bPAC input pulse sequence (max amplitude). (A) Small emitter cluster experiment with gap-junction inhibition (carbenoxolone): Emitter-receiver coupling is lost. (B) Small emitter cluster experiment with PKA inhibition (H89, 10 μM): Emitter-receiver coupling is inhibited. (C and D) Testing how Epac and PKA inhibitors affect ERK under bPAC activation for small-emitter cluster experiments. Cells are imaged during a pulsed bPAC input sequence (40 minutes off, 40 minutes on, 40 minutes off). An inhibitor is then added and the bPAC input sequence is repeated while the same cells are imaged. (C) Small emitter cluster driven by a sequence of bPAC input pulses. PKA inhibitor (H89, 10 μM) is added half way through. (D) Small emitter cluster driven by a sequence of bPAC input pulses. Epac inhibitor (ESI-09) is added half way through. (E) Experimental results for small emitter cluster surrounded by receivers that do not form gap junctions (MDCKII cells). Left panel: Image of small emitter cluster where receivers do not have ERK-KTR but have a nuclear marker (red nuclei, yellow label). Panels to the right of image: Average emitter ERK-KTR N/C signal and single cell ERK-KTR N/C signals for the small emitter cluster. (F) Scatter plots of time-to-max times, time-to-ss times, and overshoot ratio for single cell ERK-KTR N/C signals taken from the same all-emitter experiments in Fig 3D, small-emitter-cluster experiments (surrounded by receivers that do not form gap junctions), and single-emitter-cluster experiments (surrounded by receivers that do not form gap junctions). Mean (magenta diamond) ± standard deviation (magenta error bars), error bars in the mean (green, standard deviation/) represent 95 percent confidence interval in the mean, and median (yellow diamond) are also plotted. Number of data points N used are displayed below each group label.

https://doi.org/10.1371/journal.pcbi.1009873.g004

Unexpectedly, when we add PKA inhibitor (H89, 10 μM), the cAMP coupling from emitters to receivers is greatly reduced (Fig 4B (average signals) and S4(C) Fig (mean with error bars)), suggesting gap junction activity is related to PKA. Indeed, recent work [40, 41] found that PKA phosphorylates gap junctions, as an Ezrin/PKA/gap-junction (connexin 43) complex, through cAMP activation, thereby controlling gap-junction activity (permeability). Thus, PKA inhibitors are inhibitors of gap junction activity. In addition, we also observe that the average emitter ERK-KTR N/C signal did not exhibit fast, dynamic overshoot (Fig 4B), thus, strengthening support for our hypothesis that gap-junctions are responsible for the strong, fast overshoot observed in many emitters from small emitter clusters.

While both PKA and Epac inhibit ERK [17, 18] and thus both contribute to the ERK-KTR N/C signal during cAMP induction through bPAC (see S3 Text and S5(A) and S5(B) Fig), Epac is not known to inhibit gap-junctions. Consider a small-emitter-cluster, two-pulse bPAC-input experiment where an inhibitor is added halfway between the two pulses. Our results from Fig 4B, predict that during the second pulse of a PKA inhibition experiment of this type, the emitters will show little to no dynamic overshoot in the ERK-KTR N/C signal as well as negligible coupling to receivers, which is what we observe (Fig 4C). However, for an Epac inhibition experiment the emitter should still be capable of exhibiting dynamic overshoot as well as coupling to receivers, which is what we observe (Fig 4D), providing further evidence of gap junctions being the culprit causing the observed overshoot.

As a final control, these cell-cell coupling insights make the prediction that small emitter clusters and single emitter clusters that are surrounded by receivers that do not form gap junctions (MDCKII cell line) should exhibit emitter ERK-KTR N/C signals that resemble all-emitter monolayers. Indeed that is what occurs in both small emitter clusters (Fig 4E and S4(D) Fig) and single emitter clusters (S4(E) Fig). As expected, the time-to-max statistics also shift upwards to the all-emitter regime (Fig 4F, left panel, small-emitter-cluster and single-emitter-cluster data, mean error bars in green). Statistically, the pooled data of the small emitter clusters surrounded by receivers with no gap junctions has a greater mean than the pooled small-emitter-cluster data from Fig 3D (left panel) (p < 10⁻⁵ (see Materials and methods for details)). In addition, as might be expected both time-to-ss and overshoot ratio for the small emitter clusters and single emitter clusters surrounded by receivers that do not form gap junctions behave similarly to the all-emitter data (Fig 4F, middle panel (time-to-ss) and right panel (overshoot ratio), small-emitter-cluster and single-emitter-cluster data). Statistically, pooled small-emitter-cluster data from Fig 3D (right panel) has a greater mean for the overshoot-ratio than the data for the pooled small emitter clusters with receivers that have no gap junctions (p < .012 (see Materials and methods for details)).

Altogether, these experimental results establish that the coupling of cAMP from emitters to receivers through gap-junctions is what gives rise to the overshoot dynamics observed in many of the emitters in the small-emitter-cluster and single-emitter-cluster experiments.

A simple model of the all-emitter experiments

The ERK-KTR N/C signal overshoot dynamics that can be observed in emitters of small-emitter-cluster and single-emitter-cluster experiments, but not for emitters in all-emitter experiments, inspired us to develop a basic model that can account for these observations.

We first wanted to construct a simple intracellular model based on the experimental observations above. To summarize, the rise and decay dynamics of the emitter ERK-KTR N/C signal are very similar between all-emitter experiments (Fig 2E and 2F, for example) and small-emitter-cluster experiments surrounded by receivers with no gap junctions (Fig 4E and S4(D) and S4(E) Fig). They all follow simple first-order like dynamics. Furthermore, if, for example, we apply a pulse sequence of increasing amplitude to an all-emitter monolayer with PKA inhibition (H89) (S5(C) Fig), which has reduced cell-cell coupling, we observe a similar response in the ERK-KTR N/C signal as that for the a similar no-drug all-emitter experiment (Fig 2F). The time-to-max statistics are similar as well (S2(E) Fig). All these observations suggest that we are observing how the intracellular regulatory circuit sculpts the ERK-KTR N/C signal. And even though there is cell-cell coupling for the wild-type all-emitter experiments, there should be no strong net-flux, on average, between cells, unlike the emitter-receiver coupling in the wild-type small-emitter-cluster experiments. This is because all cells are producing cAMP in the all-emitter experiments.

In constructing a simple intracellular circuit, we consider known feedbacks on cAMP (see intracellular circuit in S5(D) Fig, a literature-based model with feedbacks on PDE from [42]), including PKA and Calmodulin Kinase II feedback. Because rise and decay dynamics of the ERK-KTR N/C signal are very similar for the PKA inhibited all-emitter experiments and the wild-type all-emitter experiments, this suggests that PKA might not have strong feedback through PDEs on cAMP. In addition, recent work [42] has uncovered a feedback through Epac to Calmodulin Kinase II to PDE4. The bPAC induced all-emitter experiment that inhibits Epac halfway through shows a similar time-dependent shape of the ERK-KTR N/C signal before and during Epac inhibition (S5(B) Fig). It is possible that enough Epac might be active during inhibition to still provide negative feedback through Calmodulin Kinase II onto to PDEs to ensure this. Or it is possible that the feedback through Calmodulin Kinase II is not significant in the regime we are testing the system.

Given that there is not much difference in the ERK-KTR N/C signal dynamics with or without the inhibitors (PKA or Epac), we will consider a more simple intracellular circuit where the cAMP degradation is driven by a simple rate constant, γ_pde. For this case, the cAMP degradation term is simply [cAMP]γ_pde, the concentration multiplied by the rate constant, i.e. a simple first order decay (Fig 5A, intracellular circuit, see S4 Text for further details of the computational model).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Model requires gap junction regulation with a delay to recapitulate difference in ERK-KTR N/C signals between emitter and adjacent receiver cells.

(A) Illustration of simplified intracellular pathway. (B-E) Modeling results for simplified intracellular pathway with no feedbacks in an all-emitter monolayer. Model is fitted to ERK-KTR N/C data for 6 minute bPAC input pulses applied to an all-emitter monolayer (S1(B) Fig, IBMX, population average). ERK-KTR N/C and cAMP signals shifted to zero and scaled for comparison, with the peak of the emitter signal at 1 and minimum at zero. (B) ERK-KTR N/C signal and predicted cAMP signal driven by 6 minute bPAC input pulse. We present results for two values of the deactivation rate of ERK γ_erk using the expression κ_erkγ_erk where κ_erk is a scaling factor and γ_erk is a fitted deactivation rate. Top plot: fitted model (κ_erk = 1). Middle plot: same fitted model except for a slower ERK deactivation rate reduced by a factor of 2 (κ_erk = .5). (C) Fitted model results (κ_erk = 1) for ERK-KTR N/C signal and predicted cAMP signal, with and without IBMX. (D) Predicted 40 minute bPAC input pulse driven ERK-KTR N/C and cAMP signals (with and without IBMX). Signals shifted to zero and normalized for comparison. (E) Modeled cAMP signal (top plot), and with PDE inhibitor (IBMX, middle plot). Same signals as in (D) but not zero shifted and normalized. (F) Multicellular system to be modeled with 1 emitter (yellow nucleus, red label) and receiver cells (light green nuclei, black label). Lines connecting nuclei represent direct cell-cell communication through gap-junctions. (G-I) Single-emitter-cluster modeling using system in (F): For top two plots, the ERK-KTR N/C and cAMP signals are shifted to zero and scaled for comparison, with the peak of the emitter signal at 1 and the minimum at zero. The receiver signals are multiplied by the same scaling factor as the emitter to maintain the same relative amplitudes (proportionally). (G) Results for the simplified intracellular pathway with constant gap junctional permeability. (H) Results for the intracellular PKA feedback model through PDE that produces overshoot (in all-emitter monolayer) with constant gap junctional permeability. (I) Results for the simplified intracellular pathway with cAMP/PKA regulation of gap junctions with a delay. Corresponding dynamics of the active gap junctions on the emitter/receiver interfaces (averaged) is also included with a breakdown of the cAMP/PKA driven contributions from the emitter and the average of the adjacent receivers.

https://doi.org/10.1371/journal.pcbi.1009873.g005

We model the ERK-KTR N/C signal as a two step process. First, the combined effects of cAMP driving both Epac and PKA to deactivate ERK (through Rap1 [10]) are represented by a simple cAMP-dependent deactivation rate of ERK. Second, the effect of ERK driving EKT-KTR into the cytosol is represented as simple ERK-dependent, cytosolic translocation rate of ERK-KTR (details in S4 Text). The model was fit to data from the 6 minute bPAC input pulse experiments discussed in S1 Text (see S6(A)–S6(C) Fig for model/fit). We present two fitted examples whose ERK deactivation rate differs by a factor of two (Fig 5B and 5C). Both cases show a slightly delayed ERK-KTR N/C response relative to the modeled cAMP signal, with the slowest ERK deactivation rate (κ_erk = .5) case exhibiting the most delay. While a little sluggish relative to the cAMP signal, we will show that the fitted ERK-KTR N/C signals mirror the salient qualitative features of the cAMP signal such as overshoot. We also simulate the system in the presence of PDE inhibitor (IBMX) which expectedly decreased the decay rates (Fig 5C, IBMX cases) of the signals upon turnoff.

Results for this fitted simple intracellular model, including the model of the ERK-KTR N/C signal, are presented for an all-emitter monolayer for a 40 minute pulsed bPAC input which captures the basic first order rise and turn off of the cAMP signal and ERK-KTR N/C signal (Fig 5D and 5E and S6(D) and S6(E) Fig, with and without the PDE inhibitor IBMX) as well as PKA (S6(F) Fig). With this model of the intracellular circuit, we turn our focus to modeling the cell-cell coupling mechanism which we deduced to be the major factor for the strong overshoot of the ERK-KTR N/C signal observed in the small-emitter-cluster and single-emitter-cluster experiments.

Determining a simple regulated gap-junction flux model for the cell-cell coupling of cAMP

The data measured from the experiments presented in Figs 2–4 implicates gap-junctions as the mechanism for the overshoot observed in the ERK-KTR N/C signal in single emitters from single-emitter-cluster experiments. However, our experiments do not measure the gap-junction activity at cell-cell interfaces that would drive this phenotype. Here we explore computational gap-junction models that could support our experimental observations. As a foundation, we apply a standard gap-junction flux model [43, 44] where the model for the flux of cAMP through gap-junctions at the interface of cell i and cell j is (1) Here ρ_ij is the effective gap-junctional cAMP permeability (can be time-dependent) at the interface of cell i and cell j. And [cAMP^j] − [cAMPⁱ] is the difference of cAMP concentrations at the gap-junction interface and dictates the direction and magnitude of the flux.

We first applied a simple cell-cell coupling model of constant gap junctional permeability, i.e. the coupling capability is constant over time. For this case we set ρ_ij = ω_ijk_gj, where ω_ij = 1 when cells i and j interface, otherwise ω_ij = 0. And k_gj is a constant resulting in constant gap junctional permeability across all cell-cell interfaces. Thus for this case, ρ_ij is not time-dependent. For a single emitter surrounded by receivers pictured in Fig 5F, we find that this model cannot reproduce the strong overshoot (Fig 5G and S6(G) Fig, modeled over a range of constant gap junctional permeability values to test different coupling strengths). Next we tested a strong intracellular PKA feedback through the PDEs that is capable of producing dynamic overshoot in an all-emitter monolayer (S6(I) Fig). While we already know it is not representative of the actual intracellular system, we wanted to see how the receiver cells responded. When we run the single-emitter-cluster case from Fig 5F, we get the expected overshoot in the emitter, but the receiver mirrors the emitter with overshoot (Fig 5H and S6(H) Fig, modeled over a range of constant gap junctional permeability values to test different coupling strengths), not the simple response with no overshoot observed by the adjacent receivers in the single-emitter-cluster experiment in Fig 3E. Thus, this simple cell-cell coupling model of constant gap junctional permeability is insufficient.

From a regulation perspective, the simple gap junction model imposes a negative regulation on the single emitter intracellular cAMP signal since cAMP is leaving the emitter through gap junctions to the surrounding receivers. However, this is a time independent regulation. It is well known, especially in feedback systems, that delayed negative regulation on a signal is a mechanism that can cause overshoot in that signal [45]. With this in mind, we hypothesized the overshoot to be due to cAMP/PKA dependent gap-junction regulation with a delay. When cAMP levels go up there is an increase in gap-junction permeability with about a 10 minute delay to reach a maximum permeability, potentially related to either transport of connexin 43 onto the membrane and/or the formation of additional PKA/ezrin complexes tethered to connexin 43. At about 10 minutes this ‘opens the flood gates’ enabling greater flux between cells thus allowing for a strong overshoot in the emitter cells. A simple means of capturing this effect involves a series of finite delay steps initiated by cAMP/PKA to activate gap-junction molecules. Here, is the inactive gap junction molecule, in the ith cell of the monolayer, which goes through a series of N intermediate delay steps where represents the gap junction molecule at the kth delay step, for k = 1,2,‥,N. After the final delay step the molecule is then in the active state, , a fully formed active gap junction. This is captured by the chemical equations (2) where . Here τ_gj is the mean time it takes for a group of K gap-junction molecules in the first delay state to reach the active state . Here the mean delay τ_gj could represent transport and/or Ezrin/PKA/connexin complex formation prior to becoming active and coupling to a neighboring cell. For N = 1 steps, the time dependent accumulation of the group into the active state can be expressed as K(1 − exp(−t/τ_gj)), i.e. highly dispersed in activation times. For large N, it is a switch like (perfect delay) response, Ku(t − τ_gj), a step function where every molecule becomes active at the same time (no dispersion in activation times). For both cases we are assuming the group starts in state at t = 0. Intermediate values of N will have intermediate dispersion of activation times relative to the two extremes presented (see S6(J) Fig for examples). For those molecules in the active state, they can then de-activate (endocytosis, for example) back to the inactive state via the chemical equation (3)

The delayed regulation model in cell i and in cell j modulates the effective gap junctional cAMP permeability, ρ_ij, where the flux now has the form (4) where ω_ij = 1 when cells i and j interface, otherwise ω_ij = 0. Here is the active concentration initiated by the delayed regulation model in cell i and likewise in cell j. P_i is the number of cells that cell i couples to and P_j is the number of cells that cell j couples to. Within each cell, the active concentration is parsed equally to each cell it couples to, hence are the parsed concentrations from cells i and j, on the ij interface. Thus, ρ_ij is time-dependent and modulated by cAMP/PKA activity. The term is the average number of cells that the cells couple to. In this work we set , where most P_i typically vary between 4 and 7. and are normalizing constants so that k_gj,gjf has the same dimensions as the basal term k_gj. For further details, including the ordinary differential equations (ODEs) of the delayed regulation model, the cell-cell coupling flux model , and all molecular species of the multicellular model, please see S4 Text. To reiterate, with this approach, our multicellular model calculates the time-dependent active gap-junction concentrations, and hence the time-dependent effective gap junctional cAMP permeability at all cell-cell interfaces. We are therefore able to calculate all time-dependent cAMP fluxes between cells through the gap-junctions.

Implementing this approach, we find the simple intracellular model with delayed gap-junction regulation is able to capture both the strong overshoot of the ERK-KTR N/C response in the emitter and the simple response with no overshoot in the adjacent receivers (Fig 5I, N = 4 and τ_gj = 15, for both the fitted ERK-KTR models (κ_erk = 1 and κ_erk = .5 (slowest))). Corresponding dynamics of the active gap junctions on the emitter/receiver interfaces (averaged) is also presented in Fig 5I with a breakdown of the contributions from the emitter and the average of the receivers. Furthermore, adjusting N and τ_gj modulates the overshoot behavior (S7(A) Fig). And decreasing the number of total receiver cells the single emitter couples to, from 45 cells (Fig 5F) to 10 cells (S7(C) Fig), still results in emitter overshoot while marginally increasing the amplitude of the receiver signal (S7(C) Fig).

We found experimental evidence that supports our delayed gap-junction regulation model. In particular, studies of fluorescent dye flux through gap junctions [46] show that upon titrating in 8-bromo-cAMP, cells exhibit both an increase of flux through connexin43 gap junctions (Figs 1 and 2 from [46], where connexin43 gap junctions are the most efficient gap junction transporters of cAMP [47]) and an increase in membrane gap-junction plaques by 15 minutes (Fig 6 from [46]). This delay in the increase of plaques, which represent regions of active coupling channels, and increased flux is consistent with our cAMP/PKA-dependent, delayed gap-junction regulation model.

Model predictions and experimental validation

Our model predicts that we will observe similar ERK-KTR N/C overshoot behavior in the presence of the PDE inhibitor IBMX (S8(A) Fig, right panel) as we did without (Fig 5I), and with similar active gap junction molecule dynamics (compare S9(A) Fig (IBMX) with S9(B) Fig (no drug)). The result holds over a range of parameters (S7(B) Fig). We confirmed this experimentally for a small emitter cluster in the presence of the PDE inhibitor IBMX (S8(A) Fig, left panel).

Likewise, the model predicts for the small-emitter-cluster case that a reduction of available gap-junctions to couple to receiver cells will make the emitter signals look more like the all-emitter case (S8(B) Fig, lower right panel (model)). This makes sense since less cAMP flux is entering receiver cells. Indeed this is the case. For a receiver cell that has over expressed connexin-43 with a GFP tagged to its N-terminus (C43-NGFP) [48], these connexins can form gap-junctions at the membrane with other cells, but cannot open [49]. We denote these receiver cells as receiver-CX43-NGFP cells. And its connexin population therefore reduces the level of endogenous working gap-junction channels, thereby reducing flux (upper illustrations in S8(B) Fig). Thus, emitters in a small emitter cluster coupled to receiver-CX43-NGFP cells cause the emitter cells to behave more like emitters in the all-emitter case (S8(B) Fig, lower left panel), with much less overshoot in the ERK-KTR N/C signal.

Furthermore, for small emitter clusters coupled to receiver cells, if we add an inhibitor of gap junction molecule (connexin) trafficking from the ER to the golgi (Brefeldin A), after 4 hours we should see less overshoot in the ERK-KTR N/C signal in the emitters as well as higher amplitude. This is because connexin trafficking from the golgi to the membrane is greatly reduced [50] because of a lack of connexins in the golgi. In addition, membrane gap-junctions get endocytosed thus depleting the existing membrane gap-junction population as time proceeds. Indeed we see reduced overshoot and increased ERK-KTR amplitude when bPAC is on (S8(C) Fig), with low levels of coupling to receivers.

Overall, we have been able to predict with our model and experimentally confirm how different chemical and genetic perturbations to cell-cell coupling affect the ERK-KTR N/C signal in emitters and adjacent receivers.

Strong heterogeneity in ERK-KTR N/C response in emitters of single-emitter-cluster experiments indicates a dynamically changing gap junction state

For emitters in an all-emitter experiment, our variability analysis (see S2 Text) shows that the ERK-KTR N/C response tends to be very similar within a given emitter across bPAC input pulses of the same amplitude (single cell responses in S2(A) Fig, for example). In addition, emitters in the all-emitter experiments generally exhibit a very simple first-order like behavior in the responses. However, from our single-emitter-cluster experiment in Fig 3E, it is interesting that there is such a striking difference in the single emitter response across bPAC input pulses of the same amplitude. Indeed, for single-emitter-cluster experiments driven by bPAC input pulses of the same amplitude, we identified that the single emitters tend to exhibit a much larger variation in the dynamics of the ERK-KTR N/C response to bPAC both between different experiments and across different pulses within a given experiment (Fig 6A and S3(A) Fig (right panel)). For example, some overshoots are faster or have higher amplitudes than others, or there may be no overshoot during one of the pulses or both. Furthermore, while our pooled single-emitter-cluster experiments show lower time-to-max statistic for single emitters, on average, than emitters in the all-emitter experiments (Fig 3D, left panel), there are numerous outliers that are much higher (representative example in Fig 6A, second emitter plot from bottom, response during first pulse).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Differences in the emitter ERK-KTR N/C response across bPAC input pulses of the same amplitude reveal the influence of a dynamically changing gap junction state in single-emitter-cluster experiments.

(A) Single cell ERK-KTR N/C signals from emitters of different single-emitter-cluster experiments all driven by two-pulse bPAC inputs of the same amplitude. (B) Illustration of the fast gap-junction population (black, near the plasma membrane) and the slow gap-junction population (red, closer to the nucleus). (C) Modeling of gap-junction regulation heterogeneity: Case 1 (far only), Case 2 (near + far), Case 3 (near only). Results are presented for both of the fitted ERK-KTR models (κ_erk = 1 (blue) and κ_erk = .5 (red, slowest), as was done in Fig 5I). (D) Modeling the undershoot observed in some single emitters. Top plot: gap-junction regulation only. Middle plot: gap-junction regulation and endogenous adenylyl cyclase feedback. Bottom plot: bPAC input. (E) Modeling the undershoot observed in some single emitters in the presence of PDE inhibitor (IBMX). Top plot: gap-junction regulation only. Middle plot: gap-junction regulation and endogenous adenylyl cyclase feedback. Bottom plot: bPAC input.

https://doi.org/10.1371/journal.pcbi.1009873.g006

To model this observed variability, we altered our existing model of cAMP/PKA regulated gap-junction activity. Our initial model had particular initial levels of membrane bound gap-junctions and levels of gap-junctions in the recycling/trafficking pathway (see ‘far population’ in Fig 6B representing inactive gap-junction molecules (connexins) near the nucleus/golgi that must be transported resulting in a large delay (∼10 to 15 min.) before activation on the membrane). To model the variability observed in the ERK-KTR N/C response across cells or different pulses within a cell, we add another population, a near population, representing inactive gap-junction molecules (connexins) at or near the plasma membrane that has a smaller activation delay than the far golgi-located population (see ‘near population’ in Fig 6B for illustration). One could imagine a continuum of populations, each with a different delay, but for this analysis we will consider only two.

Variability in the modeled cAMP and ERK-KTR N/C signals are obtained by adjusting the initial levels of these 2 populations of inactive gap-junctions (see Fig 6B for illustration) for each pulse within a cell. The motivation here is that the recycling/trafficking pathway, during a pulse, alters these populations as well as the active population on the membrane. And the populations could further get altered by the recycling/trafficking pathway in between pulses. Thus, at the beginning of the next pulse, the cell will likely have different initial conditions than it had for the prior pulse. For our initial simulations, the near population gets up-regulated within 1 minute or so (parameters: τ_gj = 1, and N = 4), while the far population gets up-regulated strongly by 15 minutes (parameters: τ_gj = 15, and N = 4). For all simulation runs the total initial amount of inactive gap-junction molecules we set at N_i, where , i.e. the sum of the two populations equals N_i. Depending on the amount in each population, we can obtain different cAMP response overshoot signatures, the far only case (Fig 6C, case 1, ) being similar to the data from Fig 6A, third emitter plot from top (first pulse) and bottom emitter plot (second pulse). The near + far modeled response (Fig 6C, case 2, ) and near only modeled response (Fig 6C, case 3, ) can be seen in experimental data in the top emitter plot of Fig 6C with a ‘near + far’ type response during the first pulse and a ‘near’ type response during the second pulse. Results in cases 1–3 are presented in Fig 6C for both of the fitted ERK-KTR models (κ_erk = 1 (blue) and κ_erk = .5 (red, slowest), as was done in Fig 5I). The corresponding active gap-junction activity at the emitter/receivers interface for cases 1–3 are presented in S9(B)–S9(D) Fig.

These experimental and modeling results suggest a dynamically changing gap-junction state, i.e., a dynamic gap junctional cAMP permeability, in single emitters across pulses that is revealed most strongly in single-emitter-cluster experiments relative to the all-emitter experiments. Importantly, we also see evidence of approximate constant gap junction activity (Fig 6A, second emitter plot from bottom) where, especially for the second pulse, the response has no overshoot and is flat similar to our model of constant gap-junction activity (Fig 5G). And during the first pulse there is a slow continuous rise which could be due to a net loss of gap-junctions over time through endocytosis.

Going back to the single-emitter-cluster example from Fig 3E, one of the receiver cells undergoes mitosis a little before the second bPAC input pulse begins. It has been observed in endothelial cells that during mitosis, gap junctions are removed from the membranes and are intracellular, leading to a loss of gap junction intercellular communication with surrounding cells [51]. Surrounding cells that are non-mitotic keep their gap junctions at cell-cell interfaces and still communicate. Since this loss of gap junction function is likely happening with the receiver cell undergoing mitosis (Fig 3E, cell 37), it would no longer receive cAMP from the emitter cell. Thus, the overshoot in the ERK-KTR N/C signal during the second bPAC input pulse would be due to the remodeling of the gap-junction interfaces in the adjacent non-mitotic receivers and the emitter.

All of these different single cell results point to a picture of the gap-junction state being highly dynamic and heterogeneous between different single-emitter-cluster experiments and across time within a single emitter.

Cell culture and reagents

MDCKI and MDCKII cell lines were a kind gift from Pavel Nedvestky. No mycoplasma contamination was found. MDCKI + bPAC cells were generated by infecting MDCKI cells with pLenti-PGK- bPAC::NES (Addgene #130267). ERK-KTR expressing cells were obtained by transfecting MDCKI and MDCKI + bPAC cells with either pPB-CAG-RFP-ERKKTR or pPB-YFP-ERKKTR. PKAc (PRKACA reporter) expressing cells were obtained by transfecting MDCKI and MDCKI + bPAC cells with pPB-CAG-PRKACA::mRuby2 (Addgene #130268).

Overexpressed CX43-NGFP (broken connexin43) expressing cells were obtained by transfecting MDCKI and MDCKI + bPAC cells with sfGFP-Cx43K258stop (Addgene #69025) or msfGFP-Cx43 (Addgene: #69024).

Plasmids for the overexpressed CX43-NGFP (broken connexin43) expressing cells were constructed using the Mammalian Toolkit (MTK) [53], a hierarchical DNA assembly method based on Golden-Gate (GG) cloning [54]. New binding domains were domesticated into the MTK by ordering primers or gBlocks from IDT that contained the desired coding sequence with all BsaI and BsmBI restriction sites replaced with synonymous mutations and overhangs for MTK Part 3as, then ligated into the MTK part entry vector using a BsmBI GG reaction. The plasmids were added then to MDCKI + bPAC + RFP-ERKKTR and MDCKI + RFP-ERKKTR cells for experiments when ERK-KTR measurements were also required.

Cells were maintained in MEM (Gibco #11095072) supplemented with 10 percent Fetal Calf Serum (VWR #89510–184) and 1X Anti-Anti (Gibco #15240062), and kept at 37 degrees (C) in a humidified incubator with 5 percent CO2.

For activators and inhibitor experiments, small molecules and doxycycline were dissolved to typically 4X to 10X greater than the IC50 concentration (EC50 for activation) in imaging media before addition. The aim was for the effect of the drug to be maximized (either inhibition or activition) while minimizing stress and other physiological effects, when possible. Final DMSO concentration did not exceed 0.15%. Inhibitors included Epac inhibitor ESI-09 (15 μM, ESI-09 IC50 inhibition of Epac1 at 2.4 μM and Epac2 at 4.4 μM, in presence of 20 μM cAMP (competitive binder)), PKA inhibtor H89 (10 μM, IC50 of .135 μM), PDE inibitor IBMX (100 μM, IC50 values are 13, 18, 19, 32 and 50 μM for PDE4, PDE3, PDE1, PDE5 and PDE2 respectively), Gap junction inhibitor carbenoxolone (10 μM), ER to GolgI trafficking inhibitor Brefeldin A (10 μM). Activators included the adenalyl cyclase activator Forskolin (50 μM, saturating). All inhibitors and activators were purchased from Tocris.

Microscopy

To prep cells for continuous experiments (including those when drugs were added prior to imaging), cells were plated in Greiner Bio-One 4-compartment dishes (GBO #627870) where each compartment received 1 mL of cells/media (2.5*10⁵ cells/mL with media phenol-red-free MEM (Gibco #51200038) with 2% Fetal Calf Serum (VWR #89510–184)). To prep cells for experiments when drugs were added during imaging, cells were plated in Ibidi μ-Slide 8 Well Grid-500 (Ibidi #80826-G500, .9*10⁵ cells/mL, .3 mL of cells/media per well). This prep was also used for experiments involving the PRKACA reporter that quantifies PKA activity. For any of the above experiments, the plated cells were then incubated for at least 16 hours prior to imaging. An hour before imaging, the cells were stained with DNA binding dyes (Hoechst) as a nuclear marker (DAPI channel). The same media phenol-red-free MEM with 2% Fetal Calf Serum was replenished and if any small molecule was required at the beginning of the experiment, it was added with the replenished media. Imaging would begin at least an hour later.

Widefield microscopy images were taken on a Nikon Ti-E inverted scope, with an incubation enclosure set to 37 C (Ibidi Temperature controller), 5.0 percent CO2 and 80 percent humuidity (Ibidi CO2 and humidity controller), and mercury arc-lamp illumination using RFP filters and dichroic mirror from the Chroma 89006 ET-ECFP/EYFP/mCherry filter set. The microscope’s perfect focus system was used to maintain focus on the samples throughout the experiment. The microscope was controlled with micro-manager software version 1.4.17 using the High Content Screening Site Generator Plugin (Edelstein et al., 2014). We used custom MATLAB scripts to control micro-manager and blue LED/Halogen lamps with a neutral density filter (max light intensity of F_max ≈ 200 μW cm⁻² where the EC50 of bPAC activation saturation is 370 μW cm⁻² [22].) to induce emitter cells (cells containing bPAC) with user-defined blue light inputs. Note that for the 6-minute-pulse experiments in S1(B) Fig, the neutral density filter was not used and α represents the resulting scaled increase in photon flux where α ≈ 7. Images were taken with an Andor 512 pixel EMCCD camera (897 iAxone DU-897E) using a 40x objective (Nikon Planfluor 40x/0.75) and 1.5x optical zoom. This resulted in a typical range of 40 to 80 cells in the field of view during acquisition. At 40x, the camera resolution is .4 microns per pixel (example distance calculation in S3(D) Fig).

Key conditions from imaging experiments were performed at least twice, with one replicate presented in figures.

Choice of 40 minute bPAC input pulses

For the majority of our experiments we use 40 minute bPAC inputs pulses. The 40 minute duration was chosen because the majority of cells reach steady state in their ERK-KTR N/C signal by that time. Thus, we can observe the single-cell dynamic response to completion. Twenty minute pulses are too short where the ERK-KTR N/C signal can still be exhibiting dynamics by twenty minutes. In addition, we’ve done 8 hour experiments on single monolayers with duty cycle 40 minute on/40 minute off, and the cells appear healthy during the whole experiment. And for a similar reason we chose an off time of 40 minutes so the cell tend to be at an approximate steady-state when the next bPAC input pulse starts.

Image analysis and signal extraction and quantification of pathway activity reporters

We used custom MATLAB nuclear segmentation scripts that apply adaptive thresholding approaches [55] on the sequence of DAPI images obtained which results in a group of objects, some single nuclei and some clustered/overlapping nuclei (typically 2–5 nuclei). A recursive adaptive thresholding algorithm with a minimum nuclear size constraint (200 pixels for our 40X images) is then applied to each individual object in order to further separate clusters into individual nuclei. For each individual segmented nuclei object obtained, a threshold of half the peak nuclear DAPI intensity is applied to that nuclei in the DAPI image to obtain the final segmented object for the individual nuclei. Any remaining clusted/overlapping nuclei are discarded in further data processing and analysis. Our segmentation protocol identifies nuclei and its nuclei position (center of mass) for each image. Nuclei are tracked across images based on determining the closest individual nuclei between images, allowing for the tracking of individual nuclei over time. DAPI images are obtained every three minutes, a sufficient sampling period to minimize errors in individual nuclei assignment across images.

Emitter nuclei also have the genetic markers (tagBFP-nls or Lamin B1-mTFP1) in addition to the DNA binding dyes (Hoechst). For experiments containing emitters and receivers, the MATLAB scripts scan for this genetic markers (above a minimum threshold) within the segmented nuclei to determine the nuclei of emitter cells and receiver cells within the images.

With each nuclei identified, we determine the associated cytoplasmic region of each cell by following the established procedure of creating a ring around each nucleus with a half-nuclear width [32]. We then identify and track individual cells over time by mapping their nuclear locations and corresponding reporter readout from frame to frame. For each cell in a monolayer, we can therefore extract its spatial coordinates and the time dependent profile of the nuclear and cytoplasmic magnitude of its various reporters.

In order to measure ERK activity in MDCK cells, we have implemented an ERK responsive Kinase Translocation Reporters (KTR) [32]. Here, phosphorylation by ERK is transformed into a nucleocytoplasmic shuttling event of the KTR that can be measured by fluorescence microscopy. In this technology, kinase activity leads to phosphorylation of the Nuclear Localization Sequence—Nuclear Export Sequence of the designed reporter, and hence increases its cytoplasmic residence. Conversely, reduced kinase activity leads to accumulation of reporter in the nucleus. We successfully expressed a KTR for ERK (ERKKTR) fused to either Clover or tagRFP fluorescent proteins in MDCK cells. For each tracked cell, the ratio of the nuclear median signal over the cytosolic median signal was calculated for each frame. Cells that did not respond over the duration of the experiments were discarded when calculating the population averaged signal and population standard deviation of the signal for a given experiment.

In order to measure PKA activity in MDCK cells we have expressed Venus or tagRFP fused to PRKACA, one of the catalytic subunits of PKA [11]. Upon binding to cAMP, the regulatory subunits of PKA disengage from the catalytic subunits, allowing them to change their cellular distribution. This process is easily measured with the change from cytoplasmic puncta to whole cell diffuse fluorescence. Additionally, we tagged the endogenous PRKACA protein with Venus, in order to avoid the impact of extra copies of PKA catalytic subunits.

To measure the quantitative dynamics of intracellular Ca²⁺, we have expressed the genetically encoded calcium indicators (GECIs) GCaMP5g and RCaMP1h. GCaMP5 is based on a circularly permuted green fluorescent protein (cpGFP), calmodulin (CaM), and a Ca²⁺/CaM-binding (M13) peptide. RCaMP1h, on the other hand, is engineered from circular permutation of the thermostable red fluorescent protein mRuby. Upon binding to calcium, these molecules change their conformation, and by doing so, increase their fluorescence. Both sensors have high sensitivity and rapid dynamics [37, 38].

Statistical methods

We applied the student t test for paired samples (single tailed) to show that time-to-max measurements of the ERK-KTR N/C signal from one group of pooled experiments has a greater mean than that of another group of pooled experiments. We also applied the student t test for paired samples (single tailed) in the same manner for overshoot ratio measurements. For each of the pooled data sets, we used the measurements from each of the two bPAC input pulses for each single cell. The p-values are reported in the main text where the means are compared between data sets.

Computational model

The computational model representing the multicellular cAMP signaling network is described in S4 Text. Model simulations were performed with MATLAB ordinary differential equation solver ODE15S. Parameter values for the model simulations are presented in S2 and S3 Tables.