Drug-Tolerant Persister States and Tumor Regrowth

Why a Persister State Matters

A tumor can shrink during treatment yet retain a small population of cells that is less affected by the drug. If those cells occupy a reversible drug-tolerant state, their survival need not imply a permanent genetic resistance mechanism. The surviving state can become more prominent as sensitive cells are depleted, then contribute cells to the growing population after treatment stops.

The key distinction is between how many persister cells remain and what fraction of the remaining tumor they represent. A fraction can rise sharply even when the persister count stays constant or declines, simply because sensitive cells are lost more quickly. A model that tracks both cell states makes this distinction explicit and represents state switching as well as drug-specific growth and killing.

Definition: A drug-tolerant persister state is a modeled cell state with lower growth and lower drug-induced kill than the drug-sensitive state, in which cells can survive treatment and can transition back to the sensitive state. In a two-state model, \(S(t)\) and \(P(t)\) denote the numbers of drug-sensitive and persister-state tumor cells at time \(t\), respectively.

A Reversible Two-State Model

Let \(g_S\) and \(g_P\) be the per-capita growth rates of sensitive and persister cells, and let \(k_S\) and \(k_P\) be their per-capita drug-induced kill rates. Let \(a\) be the rate of transition from \(S\) to \(P\), and \(b\) the rate of transition from \(P\) to \(S\). These rates have units of inverse time. During drug exposure, a simple model is

\frac{dS}{dt}=(g_S-k_S-a)S+bP,\qquad \frac{dP}{dt}=aS+(g_P-k_P-b)P.

The terms \(aS\) and \(bP\) represent transfers between states. A transfer out of one state enters the other; switching alone therefore changes the composition, not the total cell count. The model assumes that each state has a constant per-capita rate within the period being modeled, that cells are well mixed, and that the modeled transitions and birth and death rates are nonnegative. Drug exposure is represented through the kill rates, which may change when treatment is withdrawn.

The persister-state assumption is typically \(g_P<g_S\) and \(k_P<k_S\) during treatment. This does not guarantee that \(P\) grows: its net rate also reflects its own growth, its kill rate, and exit to \(S\). Likewise, sensitive cells can decline while receiving treatment even if they continue to divide, provided their kill rate exceeds their growth rate and any additional effects such as switching.

Define the net rates before switching as \(u_S=g_S-k_S\) and \(u_P=g_P-k_P\). The total tumor-cell count \(T=S+P\) then changes according to

\frac{dT}{dt}=u_SS+u_PP.

The switching terms cancel. Thus, the total burden is determined by the state-specific net rates and the numbers of cells in each state, while switching affects those numbers and their proportions.

Why the Persister Fraction Can Rise

The persister fraction is \(q=P/(S+P)\), defined when \(T>0\). Differentiating this ratio and substituting the model gives a useful expression for its instantaneous change:

\frac{dq}{dt} =\frac{1}{T}\frac{dP}{dt}-\frac{P}{T^2}\frac{dT}{dt} =a(1-q)-bq+q(1-q)(u_P-u_S).

The first term increases the fraction when sensitive cells transition into the persister state; the second decreases it when persisters revert. The last term captures the relative net-rate effect. If \(u_P>u_S\), persisters have a relative survival or growth advantage under the current conditions, and this term raises their fraction whenever both states are present. The fraction can therefore rise even when \(dP/dt\) is negative: it is enough for sensitive cells to decline faster.

This identity is a derivation, not a separate assumption. Starting from \(q=P/T\), the quotient rule gives \(dq/dt=(T\,dP/dt-P\,dT/dt)/T^2\). Substituting \(P=qT\) and \(S=(1-q)T\) into the state equations yields the expression above. It also clarifies why a fraction alone cannot tell whether persister cells are expanding, remaining stable, or shrinking.

Worked Example: Initial Enrichment During Treatment

Consider a hypothetical tumor with \(S=200\) sensitive cells and \(P=2\) persister-state cells at treatment initiation. Suppose, in units of day\(^{-1}\), \(g_S=0.08\), \(k_S=0.50\), \(g_P=0.03\), \(k_P=0.04\), \(a=0.02\), and \(b=0.01\). These values give lower growth and lower drug-induced kill for the persister state.

The initial persister fraction is \(q=2/(200+2)=2/202=0.009901\), or about \(0.9901\%\). The net rates before switching are \(u_S=0.08-0.50=-0.42\) per day and \(u_P=0.03-0.04=-0.01\) per day. Substituting into the state equations gives

\frac{dS}{dt}=(-0.42-0.02)(200)+0.01(2)=-87.96\ \text{cells/day},

\frac{dP}{dt}=0.02(200)+(-0.01-0.01)(2)=3.96\ \text{cells/day}.

At this instant the total changes at \(-87.96+3.96=-84.00\) cells per day, so the tumor is shrinking. Yet the persister count is increasing because entry from the much larger sensitive population exceeds persister losses. The fraction derivative is approximately \(0.02372\) per day, or 2.372 percentage points per day at this instant. This is an instantaneous rate, not a prediction that the fraction will keep increasing linearly.

From Treatment Survival to Withdrawal

A low-kill state can be enriched during treatment through two distinct mechanisms: persisters may be produced by switching, and they may survive better than sensitive cells. After treatment stops, the kill terms may fall to zero as drug effect wanes. The post-withdrawal state equations should reflect that change rather than carry the on-treatment kill rates forward unchanged. Switching may also change after withdrawal; for instance, treatment-induced entry into the persister state might diminish while reversion continues.

The following example isolates one post-withdrawal mechanism. It assumes that drug killing has ceased, no new \(S\)-to-\(P\) transition occurs during the interval, and persister cells revert to \(S\). These are simplifying assumptions for a transparent calculation, not universal biological claims. Set \(a=0\), let \(b\) be constant, and let \(r_S\) and \(r_P\) be the off-drug net growth rates before switching. Then

\frac{dS}{dt}=r_SS+bP,\qquad \frac{dP}{dt}=(r_P-b)P.

Worked Example: Persister Reversion After Withdrawal

Suppose, hypothetically, that the tumor contains \(S(0)=100\) sensitive cells and \(P(0)=10\) persister cells when treatment is withdrawn. Let \(r_S=0.08\) per day, \(r_P=0.04\) per day, and \(b=0.05\) per day, with \(a=0\). Persisters have a lower off-drug net growth rate, but their reversion supplies cells to the sensitive compartment.

The persister equation has the solution \(P(t)=P(0)e^{(r_P-b)t}\). At 10 days,

P(10)=10e^{(0.04-0.05)10}=10e^{-0.1}=9.048\ \text{cells}.

To calculate \(S(10)\), solve \(dS/dt=r_SS+bP(0)e^{(r_P-b)t}\) using an integrating factor. This gives

S(t)= \begin{cases} e^{r_St}\left[S(0)+bP(0)\dfrac{e^{(r_P-b-r_S)t}-1}{r_P-b-r_S}\right], & r_P-b\ne r_S,\\ e^{r_St}\left[S(0)+bP(0)t\right], & r_P-b=r_S. \end{cases}

Substituting the values at \(t=10\) days gives

S(10)=e^{0.8}\left[100+0.05(10)\frac{e^{-0.9}-1}{-0.09}\right] =e^{0.8}(103.297)=229.9\ \text{cells}.

The total is approximately \(238.9\) cells, compared with 110 cells at withdrawal. In this hypothetical calculation, total burden rises because sensitive cells grow and receive cells through reversion. The persister compartment itself declines slightly. This illustrates how a state that survives treatment can contribute to post-treatment growth without being the fastest-growing state.

Interpreting Enrichment and Checking the Model

Tumor measurements often provide a total burden rather than direct state-specific counts. If state proportions are available, the fraction must be interpreted alongside the corresponding abundance or total burden. An increased persister fraction is evidence of a compositional change under the model, not by itself evidence that persisters multiplied or that a particular transition caused the change.

Worked Example: Correctly Calculating a Persister Fraction

Suppose a hypothetical baseline sample contains 200 sensitive cells and 2 persister-state cells. A later sample during treatment contains 8.35 sensitive cells and 2 persister-state cells. Treat these as state-specific counts for a simple comparison; they do not, on their own, identify the rates or mechanisms that produced the change.

At baseline, the persister fraction is \(2/(200+2)=2/202=0.009901\), or \(0.9901\%\). At the later measurement it is

\frac{2}{8.35+2}=\frac{2}{10.35}=0.1932,

or \(19.32\%\). The persister count is unchanged at 2, while the sensitive count falls substantially. The fraction has risen from about \(0.9901\%\) to about \(19.32\%\); this is enrichment among the remaining cells, not an increase in the persister count. The ratio of the later to baseline fractions is approximately \(0.1932/0.009901=19.52\).

When evaluating a fitted or simulated model, check that the state definitions match the data and that the rates are interpretable in the treatment and withdrawal periods. Useful checks include whether predicted cell counts remain nonnegative, whether the total burden can be reconciled with the modeled state counts, and whether the observed timing of shrinkage or regrowth is compatible with the proposed switching and net rates. In particular, a fitted persister fraction should not be treated as a direct measurement of a transition rate.

A persistent mismatch between predicted and observed state-specific counts can indicate that the two-state description is too simple, that switching or drug effects vary over time, or that measurements do not distinguish the states as assumed. The model also omits spatial structure, variable drug exposure, immune-mediated effects, and genetic heterogeneity unless these are added explicitly. With aggregate tumor measurements alone, different combinations of growth, kill, and switching may produce similar total trajectories; state-specific data or other evidence are needed to distinguish them.

How This Model Differs From Alternatives

A model with separate sensitive and permanently resistant subclones can represent heritable differences between populations. A reversible persister-state model instead represents a phenotype that cells can enter and leave. The two mechanisms can coexist, but they imply different interpretations of post-treatment behavior: return from \(P\) to \(S\) is permitted here, whereas a fixed resistant clone does not become sensitive merely by state reversion.

A single-population tumor-growth model may be adequate when there is no evidence that cell subpopulations respond differently. It cannot, however, explicitly represent enrichment of a surviving low-kill state or cell transfer between states. Conversely, adding persister compartments does not guarantee a better explanation: more parameters can be difficult to estimate from total tumor burden alone. The choice should be guided by the scientific question and the available measurements, not by model complexity alone.

Common pitfall: “The persister fraction increased, so persister cells must have proliferated.” This conclusion does not follow. The fraction \(P/(S+P)\) can rise because \(S\) declines faster than \(P\), because cells switch from \(S\) to \(P\), or both. Compare state-specific counts and the total burden before attributing enrichment to growth.

Key takeaway: A reversible persister-state model separates drug-sensitive and persister-cell counts, includes transitions in both directions, and allows drug effects to change at withdrawal. It explains how a small surviving state can become enriched and contribute to regrowth, while making clear that fraction changes alone do not identify the underlying mechanism.

Check Your Understanding

Answer each question, then compare with the solutions below.

In the two-state model, what does the term \(bP\) represent, and why does it not appear in the equation for total burden as a net gain?
If \(P\) stays constant while \(S\) decreases, what happens to \(P/(S+P)\)? Does that establish that persister cells grew?
After withdrawal, why might it be inappropriate to retain the same drug-induced kill rates used during treatment?
For \(S=90\) and \(P=10\), calculate the persister fraction. If \(S\) later falls to 40 while \(P\) remains 10, calculate the new fraction and state what changed.

Solutions. (1) \(bP\) is the rate at which persister cells revert to the sensitive state. It enters \(dS/dt\) positively and \(dP/dt\) negatively, so those contributions cancel in \(d(S+P)/dt\). (2) The fraction rises because the denominator falls while the numerator is fixed; this does not establish persister growth. (3) Withdrawal changes exposure, so the drug effect may decline or cease; the post-withdrawal kill terms should reflect the assumed exposure and pharmacology. (4) Initially, \(10/(90+10)=0.10\), or \(10\%\). Later, \(10/(40+10)=0.20\), or \(20\%\). The persister fraction doubled while the persister count stayed at 10, because the sensitive count declined.