Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Subpopulations and Stability in Microbial Communities

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A community with a rare persister phenotype is much more likely to be stable than the averaged model predicts.

desk verdict A novel asymptotic result showing rare phenotypes can stabilize competitive communities, but the key inequality q' > q is deferred to missing supplemental material and needs referee verification. read the letter →

arxiv 1908.02670 v2 pith:YP2FQAK4 submitted 2019-08-07 q-bio.PE

Pierre A. Haas , Nuno M. Oliveira , Raymond E. Goldstein This is my paper · ORCID
classification q-bio.PE MSC 92D2592D4015B52
keywords phenotypicswitchingsubpopulationstructureecologicalstabilityLotka–Volterrabacterialpersistersrandommatrixtheorymicrobiomelinear
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that the standard practice of averaging over the phenotypes of a microbial species before analyzing ecological stability gives the wrong answers. Using a competitive Lotka–Volterra model in which each species has a normal phenotype and a rare persister phenotype, the authors derive a quantitative asymptotic comparison between the full phenotype-resolving model and the corresponding averaged model. The central quantitative result is that the probability that the full model is linearly stable divided by the probability that the averaged model is linearly stable grows like $\exp[2c(q'-q)\varepsilon\sqrt{N}]$ as the number of species $N$ grows, so a rare, slow-growing phenotype has a stabilizing effect. Abundant phenotypes are instead destabilizing. The paper further shows that for large perturbations, such as antibiotic treatment, the full and averaged models frequently settle into different coexistence states.

What carries the argument

The load-bearing machinery is the matched pair of Lotka–Volterra systems: the phenotype-resolving model (5), where each species has normal cells $B$ and persister cells $C$ with switching rates $\kappa,\lambda$ and a small parameter $\varepsilon$ separating growth and competition, and the averaged model (1) defined from it by the consistency conditions (3). The argument runs entirely through the asymptotic expansion of the two Jacobians, $J_* = J_0 + \varepsilon J_1 + O(\varepsilon^2)$ and $K_* = K_0 + \varepsilon K_1 + O(\varepsilon^2)$. Because $J_0$ is block-upper-triangular, the full and averaged models share leading-order eigenvalues; the key object is the first-order correction $J_1$, which acts on $K_0$ as the negative definite matrix $-\kappa I$ and therefore stabilizes near-neutral eigenvalues more often than the averaged correction $K_1 = B_0 E \kappa/\lambda$. The binomial identity (8) turns that per-eigenvalue stabilization into the exponential ratio (9).

What would settle it

Compute the stabilization probabilities $q$ and $q'$ directly from Eq. (6) for many random parameter draws and check whether $q' > q$ uniformly; alternatively, measure the ratio $P(J_* \text{ stable})/P(K_* \text{ stable})$ in simulations of Eqs. (5) at fixed $\varepsilon$ and increasing $N$ and test whether it grows like $\exp[2c(q'-q)\varepsilon\sqrt{N}]$ rather than falling.

Watch

Extended reading notes

Core claim

The paper establishes that linear stability of microbial communities cannot be reliably inferred from models that lump subpopulations into a single averaged state. Concretely, for the phenotype-resolving model (5) with normal cells $B$ and persisters $C$ and small switching parameters, the Jacobian splits as $J_* = J_0 + \varepsilon J_1 + O(\varepsilon^2)$, while the averaged model has $K_* = K_0 + \varepsilon K_1 + O(\varepsilon^2)$; the leading-order spectra agree, but the first-order correction $J_1$ contains a negative definite term $-\kappa I$ that makes near-neutral eigenvalues of the full model more likely to be stable than those of the averaged model. Summing this per-eigenvalue stabilization over the $N-1$ non-outlier eigenvalues yields the exponential ratio in Eq. (9): the full model becomes overwhelmingly more likely to be linearly stable as $N$ grows. This contradicts the expectation, inherited from random-matrix stability arguments, that resolving more degrees of freedom should make a community less stable. For large perturbations such as antibiotic exposure, the full and averaged models also differ in which species survive, even when both are stable at the coexistence equilibrium.

Load-bearing premise

The key inequality $q' > q$, asserted after Eq. (6), is the load-bearing premise; its proof is deferred to the Supplemental Material, which is not part of the arXiv version, and if it failed the exponential stabilization in Eq. (9) would be reversed.

Editorial extensions

If this is right

  • Stability conclusions drawn from averaged species models should be rechecked when subpopulation structure is present, since resolved and averaged models disagree systematically.
  • Communities with switching to a rare phenotype such as persisters become more likely to be linearly stable as the number of species grows, relative to what the averaged model predicts.
  • Adding abundant phenotypes is destabilizing because it effectively increases the number of interacting species.
  • Large perturbations, like antibiotic treatments, can drive averaged and phenotype-resolving models to different coexistence states, so predicting microbial community resilience requires resolving subpopulations.
  • The stabilizing effect only needs generic properties of the spectral distribution, so it may apply beyond competitive interactions, although the explicit proofs in the paper are for the competitive case.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: the exponential factor $\exp[2c(q'-q)\varepsilon\sqrt{N}]$ implies that even very small persister fractions can have a disproportionately large stabilizing effect in large communities; this could be tested by varying $\varepsilon$ and $N$ in synthetic microbial communities.
  • Beyond the paper: the same stabilization mechanism may operate for any rare, slow-growing, stress-tolerant subpopulation, not only antibiotic persisters; dormant spores or slow-growing variants would be natural candidates.
  • Beyond the paper: since the proof of $q'>q$ is deferred to the Supplemental Material, a direct numerical check of the inequality over a wide parameter range would be the fastest way to confirm the mechanism.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper studies the linear stability of Lotka–Volterra models of microbial communities in which each species has multiple phenotypes. It first shows numerically that an averaged model without subpopulations can disagree with the full phenotype-resolving model about stability. It then analyzes a persister-type model with a small parameter ε, expands the Jacobians of the full and averaged models, and uses a random-matrix argument to derive the asymptotic probabilities that each model is stable. The central quantitative claim is Eq. (9): when switching to a rare phenotype, the full phenotype-resolving model is exponentially more likely to be stable than the corresponding averaged model, p′/p ~ exp[2c(q′−q)ε√N]. The paper also presents numerical simulations of large perturbations, showing that averaged and full models can predict different outcomes.

Significance. If the main result holds, it is a substantive contribution to ecological stability theory: it identifies a concrete mechanism by which subpopulation structure, specifically a rare persister phenotype, can stabilize a competitive community, partially overturning May's instability paradigm. A strength of the paper is that it works with explicit dynamical equations and a model-specific Jacobian structure rather than with a purely abstract random Jacobian, and the numerical exploration is extensive, including up to 10^8 random systems. The falsifiable prediction in Eq. (9) is clearly stated. However, the quantitative claim rests on a heuristic spectral calculation and, most importantly, on an inequality q′ > q whose proof is deferred to a Supplemental Material not included in the arXiv version. The paper's main conclusion is therefore not fully verifiable from the submitted text.

major comments (3)
  1. [After Eq. (6), Eq. (9)] The central claim p′/p ~ exp[2c(q′−q)ε√N] depends entirely on the inequality q′ > q, but the proof is not presented in the main text and the cited Supplemental Material [19] is not included in the arXiv posting. The sentence 'Eq. (6) shows that J1 acts on K0 as the negative definite matrix −κI, so q′ > q [19]' is not a proof, because J1 is a 2N×2N block matrix and does not literally operate on K0. The first-order shift of an eigenvalue of K0 under J0 + εJ1 involves the off-diagonal blocks J1_12 = −B0F and J1_21 = κI as well as the diagonal block −κI, and a standard perturbation calculation is needed to determine the sign of the correction. If q′ ≤ q, the ratio in Eq. (9) would not grow and the claimed stabilizing effect of the rare phenotype would not follow for the stated model. The supplemental derivation should be made available and its main steps summarized in an accessible way.
  2. [Eq. (8)] The binomial expression in Eq. (8) treats the N−1 non-outlier eigenvalues as independent trials with common probabilities ϖ and q. This independence and identical-distribution assumption is not established: the eigenvalues of the random matrix K0 are correlated, and the row scaling by B0 in B0E means the matrix does not have i.i.d. entries in the standard circular-law setting. The argument is explicitly heuristic, but the paper should state more clearly that Eq. (8)–(9) are heuristic asymptotic estimates rather than proven asymptotics, and either justify the independence step or provide a numerical test of the predicted exponential form beyond Fig. 2.
  3. [Fig. 2, Eq. (9)] The numerical support for the quantitative form of Eq. (9) is weak. The simulations use ε = 0.01 and N ≤ 10, so ε√N ≤ 0.032; the predicted log-ratio is 2c(q′−q)ε√N, which is only of order 0.1 if c(q′−q) is O(1). The trend in Fig. 2 is consistent with the direction of Eq. (9), but the figure cannot distinguish an exponential growth from other monotone trends at such small values of the exponent. This is not a fatal flaw, but the text should avoid claiming that Fig. 2 confirms the exponential asymptotic.
minor comments (5)
  1. [Reference [19]] The Supplemental Material is cited as '[url to be inserted]' and is absent from the arXiv version; this needs to be finalized and posted with the manuscript.
  2. [Notation after Eq. (6)] The notation 'B0E' should be defined explicitly, since B0 is a vector and the product must be understood as diag(B0)E (or equivalently elementwise multiplication); otherwise Eq. (6) and Eq. (7) are ambiguous.
  3. [Final paragraph] The closing sentence 'our conclusions apply not only to competitive, but also to more general interactions' is not demonstrated by the analysis, which relies on the specific block structure of the competitive Lotka–Volterra Jacobian; this sentence should be phrased as a conjecture or conjecture plus supporting heuristic.
  4. [Fig. 2 caption] The caption lists 'Eq. (4)', 'Eq. (5), asympt.', and 'Eq. (5), exact' only in the legend text; for reader clarity, the caption should also state which markers correspond to which model, since the main text refers to large and small markers.
  5. [Fig. 4(c)] In Fig. 4(c) the category '#' combines numerical failures and cases where different species die out; the text says these are a small number of cases, but the figure does not separate them, making the result harder to evaluate.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central asymptotic stability comparison is derived from the model equations and random-matrix spectral arguments; the deferred proof of q' > q is missing support, not circularity.

full rationale

The paper's derivation chain is not circular. The asymptotic probability p in Eq. (8) is obtained from a random-matrix spectral count (the probability that an eigenvalue of K0 lies within a distance epsilon of the imaginary axis is estimated by the circular law), a Bernoulli-sum argument over N-1 eigenvalues, and the binomial theorem. The quantity q is a genuine stabilization probability, not a fitted parameter, and the full-model probability p' is computed from the same formula with q' replacing q. The exponential ratio p'/p ~ exp[2c(q'-q) epsilon sqrt(N)] in Eq. (9) is a direct arithmetic consequence of the asserted inequality q' > q. That inequality is the only load-bearing step whose proof is not exhibited in the arXiv text: the sentence 'Eq. (6) shows that J1 acts on K0 as the negative definite matrix -kappa I, so q' > q [19]' defers all details to the Supplemental Material, which is absent from the arXiv posting. This is a genuine support gap and a correctness risk, but it is not circularity: q' and q are not defined in terms of the claimed ratio, and the authors do not relabel a fitted value as a prediction. Similarly, the consistency conditions (3) define the averaged model by matching equilibrium population sizes, births, and competition, but the stability comparison is not an identity; the full and averaged Jacobians are shown to agree only at zeroth order, and their first-order difference is exactly what the spectral analysis exploits. The numerical simulations in Fig. 2 independently confirm the trend rather than define it. No self-citation to prior author work is load-bearing; the only self-reference is to the authors' own absent supplement, which is a missing-proof issue, not a circular reduction. Score 0 for circularity.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

No new entities are invented; persisters are an empirically known phenotype. The main assumptions are the i.i.d. random matrix structure and the particular equilibrium-matching definition of the averaged model. The small parameter epsilon is a hand-chosen model input, not fitted to data.

free parameters (1)
  • epsilon = 0.01 (numerics); wild-type E. coli ~1e-5 cited from Balaban et al. 2004
    Small parameter separating persister growth/competition and switching rates from normal cells; chosen for numerical convenience, with asymptotic analysis in epsilon.
assumptions (6)
  • standard math Circular law for i.i.d. random matrices with mean and variance (Tao, Vu, Krishnapur 2010).
    Used to assert that non-outlier eigenvalues of K0 are uniformly distributed on a disk of radius r ~ sqrt(N) (Fig. 3c and surrounding text).
  • standard math Perron-Frobenius theorem for nonnegative matrices.
    Invoked to assert that the outlier eigenvalue of K0 is real.
  • standard math Eigenvalue perturbation theory (Lidskii and standard first-order perturbation) for simple eigenvalues.
    Used to write nu* = nu0 + O(epsilon) for eigenvalues of K* and J*.
  • domain assumption The parameters of K0 are random with independent entries drawn from a common distribution, so the circular law applies.
    The spectral argument leading to Eq. (8) treats the entries of K0 as i.i.d. with mean -mu0; this is an assumption about the random model ensemble that is not proven.
  • domain assumption The consistency conditions (3) define the correct 'averaged model' for comparison.
    The entire comparison between full and averaged models rests on this equilibrium-matching construction; a different averaging scheme could change the quantitative and possibly qualitative results.
  • ad hoc to paper The small-epsilon scaling in Eqs. (5) is the appropriate limit for persister dynamics.
    The separation of time scales, with C's growth and both switching rates scaled by epsilon, is introduced to make the asymptotic expansion tractable; it is motivated by biology but is an ansatz.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Subpopulations and Stability in Microbial Communities." pith.science (2026). https://pith.science/paper/YP2FQAK4

@misc{pith2026190802670,
  author       = {Pith},
  title        = {Pith review of: Subpopulations and Stability in Microbial Communities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YP2FQAK4}},
  note         = {Machine review of arXiv:1908.02670}
}
read the original abstract

In microbial communities, each species often has multiple, distinct phenotypes, but studies of ecological stability have largely ignored this subpopulation structure. Here, we show that such implicit averaging over phenotypes leads to incorrect linear stability results. We then analyze the effect of phenotypic switching in detail in an asymptotic limit and partly overturn classical stability paradigms: abundant phenotypic variation is linearly destabilizing but, surprisingly, a rare phenotype such as bacterial persisters has a stabilizing effect. Finally, we extend these results by showing how phenotypic variation modifies the stability of the system to large perturbations such as antibiotic treatments.

Figures

Figures reproduced from arXiv: 1908.02670 by the authors.

Figure 1
Figure 1. FIG. 1. Stability of communities with subpopulations. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Stability of microbial communities with phenotypic [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Eigenvalue distributions of the full and averaged mod [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Stability of microbial communities to large perturba [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [19]

    [5, 6, 20–25], for (i) a detailed descrip- tion of the selection of random systems, (ii) details of the asymptotic calculations leading to Eqs

    See Supplemental Material at [url to be inserted], which includes Refs. [5, 6, 20–25], for (i) a detailed descrip- tion of the selection of random systems, (ii) details of the asymptotic calculations leading to Eqs. (6) and (7), (iii) details of the spectral analysis leading to Eq. (9), in- cluding a note on the stability of sums of stable matrices, (iv) ...

    work page

  2. [1]

    R. M. May, Will a large complex ecosystem be stable?, Nature (London) 238, 413 (1972)

    work page 1972

  3. [2]

    Roberts, The stability of a feasible random ecosystem, Nature (London) 251, 607 (1974)

    A. Roberts, The stability of a feasible random ecosystem, Nature (London) 251, 607 (1974)

    work page 1974

  4. [3]

    Allesina and S

    S. Allesina and S. Tang, Stability criteria for complex ecosystems, Nature (London) 482, 205 (2012)

    work page 2012

  5. [4]

    Grilli, T

    J. Grilli, T. Rogers, and S. Allesina, Modularity and sta- bility in ecological communities, Nat. Commun. 7, 12031 (2016)

    work page 2016

  6. [5]

    Grilli, G

    J. Grilli, G. Barab´ as, M. J. Michalska-Smith, and S. Allesina, Higher-order competitive interactions sta- bilize dynamics in competitive network models, Nature (London) 548, 210 (2017)

    work page 2017

  7. [6]

    Butler and J

    S. Butler and J. P. O’Dwyer, Stability criteria for com- plex microbial communities, Nat. Commun. 9, 2970 (2018)

    work page 2018

  8. [7]

    C. A. Serv´ an, J. A. Capit´ an, J. Grilli, K. E. Morrison, and S. Allesina, Coexistence of many species in random ecosystems, Nat. Ecol. Evol. 2, 1237 (2018)

    work page 2018

Show all 34 references
  1. [8]

    C. A. Lozupone, J. I. Stombaugh, J. I. Gordon, J. K. Jansson, and R. Knight, Diversity, stability and resilience of the human gut microbiota, Nature (London) 489, 220 (2012)

    2012

  2. [9]

    J. J. Faith, J. L. Guruge, M. Charbonneau, S. Subra- manian, H. Seedorf, A. L. Goodman, J. C. Clemente, R. Knight, A. C. Heath, R. L. Leibel, M. Rosenbaum, and J. I. Gordon, The long-term stability of the human gut microbiota, Science 341, 1237439 (2013)

    2013

  3. [10]

    K. Z. Coyte, J. Schluter, and K. R. Foster, The ecology of the microbiome: Networks, competition, and stability, Science 350, 663 (2015)

    2015

  4. [11]

    Kussell and S

    E. Kussell and S. Leibler, Phenotypic diversity, popu- lation growth, and information in fluctuating environ- ments, Science 309, 2075 (2005)

    2005

  5. [12]

    Smits, O

    W. Smits, O. Kuipers, and J. Veening, Phenotypic vari- ation in bacteria: the role of feedback regulation, Nat. Rev. Microbiol. 4, 259 (2006)

    2006

  6. [13]

    Avery, Microbial cell individuality and the underly- ing sources of heterogeneity, Nat

    S. Avery, Microbial cell individuality and the underly- ing sources of heterogeneity, Nat. Rev. Microbiol. 4, 577 (2006)

    2006

  7. [14]

    Dubnau and R

    D. Dubnau and R. Losick, Bistability in bacteria, Mol. Microbiol. 61, 564 (2006)

    2006

  8. [15]

    M. M. Turcotte and J. M. Levine, Phenotypic plastic- ity and species coexistence, Trends Ecol. Evol. 31, 803 (2016)

    2016

  9. [16]

    J. D. Murray, in Mathematical Biology, Vol. I (Springer, Berlin, Germany, 2002) Chap. 3, pp. 79–118, 3rd ed

    2002

  10. [17]

    Dots denote tensor contractions

    Throughout, we imply elementwise multiplication of vec- tors and (rows or columns of) matrices by writing sym- bols next to each other. Dots denote tensor contractions

  11. [18]

    For N = 2, systems of this form have been considered in analyses of evolutionary stability, see e.g. R. Cressman and J. Garay, Evolutionary stability in Lotka–Volterra systems, J. Theor. Biol. 222, 233 (2003)

    2003

  12. [20]

    E. J. Hinch, in Perturbation Methods (Cambridge Univer- sity Press, Cambridge, UK, 1991) Chap. 1.6, pp. 15–18

    1991

  13. [21]

    V. B. Lidskii, Perturbation theory of non-conjugate op- erators, USSR Comp. Math. Math. Phys. 6, 73 (1966)

    1966

  14. [22]

    R. A. Horn and C. R. Johnson, in Matrix Analysis (Cam- bridge University Press, Cambridge, UK, 1985) Chap. 1.3, pp. 44–56, Chap. 2.5, pp. 100–112, and Chap. 8.2, pp. 495–503

    1985

  15. [23]

    Hardin, The competitive exclusion principle, Science 131, 1292 (1960); R

    G. Hardin, The competitive exclusion principle, Science 131, 1292 (1960); R. A. Armstrong and R. McGehee, Coexistence of species competing for shared resources, Theor. Popul. Biol. 9, 317 (1976)

    1960

  16. [24]

    Harms, E

    A. Harms, E. Maisonneuve, and K. Gerdes, Mechanisms of bacterial persistence during stress and antibiotic ex- posure, Science 354, 1390 (2016)

    2016

  17. [25]

    Kussell, R

    E. Kussell, R. Kishony, N. Q. Balaban, and S. Leibler, Bacterial persistence: A model of survival in changing environments, Genetics 169, 1807 (2005)

    2005

  18. [26]

    Maisonneuve and K

    E. Maisonneuve and K. Gerdes, Molecular mechanisms underlying bacterial persisters, Cell 157, 539 (2014)

    2014

  19. [27]

    D. I. Andersson and D. Hughes, Microbiological effects of sublethal levels of antibiotics, Nat. Rev. Microbiol. 12, 465 (2014)

    2014

  20. [28]

    J. L. Radzikowski, H. Schramke, and M. Heinemann, Bacterial persistence from a system-level perspective, Curr. Opin. Biotech. 46, 98 (2017)

    2017

  21. [29]

    N. Q. Balaban, J. Merrin, R. Chait, L. Kowalik, and S. Leibler, Bacterial persistence as a phenotypic switch, Science 305, 1622 (2004)

    2004

  22. [30]

    T. Tao, V. Vu, and M. Krishnapur, Random matrices: Universality of ESDs and the circular law, Ann. Probab. 38, 2023 (2010)

    2010

  23. [31]

    Tao, Outliers in the spectrum of i.i.d

    T. Tao, Outliers in the spectrum of i.i.d. matrices with bounded rank perturbations, Probab. Theor. Rel. 155, 231 (2013)

    2013

  24. [32]

    D. S. Maynard, C. A. Serv´ an, J. A. Capit´ an, and S. Allesina, Phenotypic variability promotes diversity and stability in competitive communities, Ecol. Lett.22, 1776 (2019)

    2019

  25. [33]

    Grilli, M

    J. Grilli, M. Adorisio, S. Suweis, G. Barab´ as, J. R. Ba- navar, S. Allesina, and A. Maritan, Feasibility and coex- istence of large ecological communities, Nat. Commun. 8, 14389 (2017)

    2017

  26. [34]

    Gibbs, J

    T. Gibbs, J. Grilli, T. Rogers, and S. Allesina, Effect of population abundances on the stability of large random ecosystems, Phys. Rev. E 98, 022410 (2018)

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.