The Shape of a Signal: Size, Duration, Frequency and Place

In a PC12 cell, one growth factor says “divide” and another says “become neuron-like”, and both say it through the same kinase, ERK. PC12 is a line of rat adrenal tumour cells that grow nerve-like processes when given nerve growth factor. If the molecule that carries both orders is the same, the difference must lie in something else about the signal. This lesson is about what that something else can be.

The idea: a signal has a shape, and the cell reads it

A cell has a limited set of signalling parts to work with and a great many things to say, so it reuses them. Calcium, cyclic AMP, ERK and NF-κB turn up in pathway after pathway, with a different result each time. A row of arrows cannot explain that. We need a new vocabulary for signaling, and its first four words are how much, how long, how often and where.

Think of a doorbell with one button. A short press, a long press and a run of quick taps are different messages sent down the same wire. A cell reads its signals in a similar way:

• How much. The size of the signal. You know this one from dose-response curves and from the all-or-none switch in signaling circuits.
• How long. Minutes, or hours.
• How often. Many signals arrive as repeated spikes, and the gap between spikes can change.
• Where. At the cell surface, on an internal membrane, or beside one target and far from another.

Reading the last three takes a decoder: a downstream part whose own timing or position makes it answer one pattern and ignore another.

How long: a part that arrives late can time a signal

The lesson on Ras and MAPK gave you the PC12 result. Epidermal growth factor (EGF) switches ERK on briefly, and the cell divides. Nerve growth factor (NGF) keeps ERK on much longer, and the cell stops dividing and grows processes. The length is what counts. PC12 cells engineered to carry extra EGF receptors keep ERK active for longer when they meet EGF, and they differentiate.

A cell has no stopwatch. What it can do is compare the signal with a delayed product of that same signal.

ERK switches on the immediate early genes you met in signaling to the nucleus, and one of them is c-fos. The c-Fos protein takes time to make, and once made it is quickly destroyed. If ERK is still active when c-Fos appears, ERK and a kinase it activates add phosphates near the C-terminal end of c-Fos, and those phosphates protect it. After a brief signal ERK is already off, so c-Fos is made and lost. After a long signal c-Fos builds up, and as a transcription factor it can switch on later genes.

Picture a clinic that calls your name twenty minutes after you sign in. Only the people still in the waiting room are seen. In the cell, the signal reaches its target by a fast route and a slow route, and the target needs both.

Leon Murphy, John Blenis and their colleagues showed this for c-Fos in fibroblasts in 2002, and later for several other early gene products. Do not assume every long signal is read this way.

How often: a decoder that holds on between spikes can count

Calcium signaling showed you that calcium often rises in brief spikes. In many cell types a stronger stimulus gives more frequent spikes of about the same height. To read that, a cell needs a part that switches on fast and off slowly.

Picture a bucket with a small hole, topped up by one cup of water at each spike. Fill it rarely and it drains between cups. Fill it often and the level climbs. For a decoder that loses its activity with a fixed half-time:

fraction left at the next spike = (1/2) ^ (gap between spikes ÷ half-time of the decoder)

With a half-time of 20 seconds, spikes a minute apart leave one eighth of the activity. Spikes 10 seconds apart leave about 70 percent to build on.

One such part is calcium/calmodulin-dependent kinase II (CaMKII), an assembly of twelve kinase subunits. Calcium-bound calmodulin switches a subunit on. An active subunit can then phosphorylate its neighbour, but only while the neighbour also holds calmodulin, and a phosphorylated subunit stays active after calcium has gone. Fast spikes catch more neighbours loaded at the same moment. In 1998 Paul De Koninck and Howard Schulman gave CaMKII fixed to a surface timed pulses of calcium and calmodulin and found that this calcium-independent activity rose with the pulse rate. That is a test-tube result.

Transcription factors can be sorted by the same rule. In 1998 Ricardo Dolmetsch, Keli Xu and Richard Lewis imposed calcium spikes of chosen spacing on a line of human T cells. Closely spaced spikes activated three transcription factors, NFAT and NF-κB among them. Widely spaced spikes activated only NF-κB. A likely reason is how fast each one switches off. NFAT leaves the nucleus within minutes once calcium falls. In B cells, NF-κB was found to stay in the nucleus for more than 16 minutes after a brief calcium spike.

Medicine uses this. The brain releases gonadotropin-releasing hormone (GnRH) in pulses. In 1978 Ernst Knobil’s laboratory showed, in monkeys that could not make their own GnRH, that one pulse an hour restored the pituitary’s release of its two reproductive hormones, and that the same hormone given continuously shut the release down. Leuprolide is a synthetic relative of GnRH given as a slow-release dose, so that its level stays steady. After a first surge the pituitary goes quiet and testosterone falls, which is the aim in prostate cancer. A drug can change the shape of a signal as well as its size.

Design problem

Two hormones raise calcium in the same cell. The first must switch on gene A and not gene B. The second must switch on gene B and not gene A. You may not add a second messenger. Design how the two calcium signals differ in time, and say what property each gene’s activator must have so that it answers only its own pattern.

Pulses from feedback: a loop with a delay keeps its own time

Where do pulses come from when the stimulus is steady? In signaling circuits you saw negative feedback pull a pathway back toward baseline. Add a delay and the correction arrives late. Think of a shower with a long pipe between tap and head: you turn the tap, nothing changes, you turn it further, and you end up too hot and then too cold. A loop like that oscillates.

NF-κB sits in such a loop. One of the genes it switches on encodes its own inhibitor, IκBα. Once IκB has been destroyed, as in Toll-like receptors and NF-κB, NF-κB enters the nucleus, new IκBα is made after a delay, and it carries NF-κB back out. If the stimulus is still present, the new IκBα is destroyed and the round can repeat, as films of single cells carrying fluorescent NF-κB have shown in some cell lines (Nelson and colleagues, 2004). In 2002 Alexander Hoffmann, Andre Levchenko, Martin Scott and David Baltimore combined a computer model with cells that lacked particular IκB proteins. They concluded that IκBα drives the oscillation and that the other two IκB proteins damp it.

p53 and MDM2 also form a negative feedback loop, and stress signals covered what Galit Lahav’s films of single cells, made in Uri Alon’s laboratory, showed in 2004: p53 arrives in pulses of about the same size, and a larger dose of radiation gives more pulses on average. The MDM2 loop does not explain them alone: a 2008 study from Lahav’s laboratory found that the pulses are driven by repeated restarts from the upstream kinases ATM and Chk2, and that a second feedback loop, through Wip1, keeps them uniform. Her laboratory later used nutlin-3, a small molecule that blocks MDM2, on a planned schedule to hold p53 steady at the height of one pulse. At moderate radiation doses, cells left to pulse recovered and returned to dividing. Cells held steady often stopped dividing for good.

Is a ring of negative feedback with a delay enough to make a clock? Michael Elowitz and Stanislas Leibler tested that by building one. In the bacterium E. coli they wired three repressors, proteins that switch a gene off, into a ring, each silencing the gene for the next. Each protein takes time to make, so every order to switch off lands late and the three chase one another around the ring. A green fluorescent readout rose and fell with a period of hours in at least 40 percent of the single cells filmed. The clock was noisy, and it ran.

Where: the same kinase means something else in another place

A message can also depend on its address. In cAMP and PKA and adaptors and scaffolds you met A-kinase anchoring proteins, which hold PKA beside particular targets. A local rise in cAMP then reaches the PKA parked nearby and may never reach PKA parked elsewhere. A receptor can likewise signal from the cell surface and again from an endosome after it has been taken inside, which is the subject of the next lesson.

A cell can also make a place on demand. In 2016 the laboratories of Michael Rosen and Ronald Vale put twelve purified proteins from the start of T-cell signalling, the pathway in antigen receptors and checkpoints, on an artificial membrane with no cell present. When a kinase phosphorylated the receptor tail, the downstream proteins gathered into clusters that behaved like liquid droplets. What holds such droplets together is a large number of weak contacts between repeated binding sites of the kind you met in protein interaction modules. The droplets also sorted what entered them. Kinases were concentrated and phosphatases were kept out, which favours a signal that stays on inside.

That sorting was shown on a glass slide. The same paper reported that, in a human T-cell line, LAT mutants that clustered more switched on more ERK, but proving that a bright spot in a living cell is a separate liquid is harder. The usual test bleaches a spot with a laser and watches the brightness return. In 2019 David McSwiggen, Mustafa Mir, Xavier Darzacq and Robert Tjian examined that test and wrote: “Even proteins that engage in stable, high-affinity binding interactions can show recovery.” Proteins that bind a fixed scaffold and let go again would look the same as a liquid. For anchored kinases, place is an established part of the signal. For droplets it is established in the test tube and still argued over in the cell.