PLOS ONE

COLLECTION REVIEW

Modeling the Justinianic Plague: Comparing

hypothesized transmission routes
Lauren A. White ID1*, Lee Mordechai1,2
1 National Socio-Environmental Synthesis Center (SESYNC), Annapolis, Maryland, United States of
America, 2 Department of History, Hebrew University of Jerusalem, Jerusalem, Israel

* [email protected]

a1111111111 Abstract
a1111111111
a1111111111 The Justinianic Plague, the first part of the earliest of the three plague pandemics, has mini-
a1111111111
mal historical documentation. Based on the limited primary sources, historians have argued
a1111111111
both for and against the "maximalist narrative" of plague, i.e. that the Justinianic Plague had
universally devastating effects throughout the Mediterranean region during the sixth century
CE. Using primary sources of one of the pandemic’s best documented outbreaks that took
place in Constantinople during 542 CE, as well as modern findings on plague etiology and
OPEN ACCESS
epidemiology, we developed a series of dynamic, compartmental models of disease to
Citation: White LA, Mordechai L (2020) Modeling explore which, if any, transmission routes of plague are feasible. Using expected parameter
the Justinianic Plague: Comparing hypothesized
transmission routes. PLoS ONE 15(4): e0231256.
values, we find that the bubonic and bubonic-pneumonic transmission routes exceed maxi-
https://doi.org/10.1371/journal.pone.0231256 malist mortality estimates and are of shorter detectable duration than described by the pri-
Editor: David A. Larsen, Syracuse University,
mary sources. When accounting for parameter uncertainty, several of the bubonic plague
UNITED STATES model configurations yielded interquartile estimates consistent with the upper end of maxi-
Published: April 30, 2020
malist estimates of mortality; however, these models had shorter detectable outbreaks than
suggested by the primary sources. The pneumonic transmission routes suggest that by
Copyright: © 2020 White, Mordechai. This is an
open access article distributed under the terms of
itself, pneumonic plague would not cause significant mortality in the city. However, our
the Creative Commons Attribution License, which global sensitivity analysis shows that predicted disease dynamics vary widely for all hypoth-
permits unrestricted use, distribution, and esized transmission routes, suggesting that regardless of its effects in Constantinople, the
reproduction in any medium, provided the original
Justinianic Plague would have likely had differential effects across urban areas around the
author and source are credited.
Mediterranean. Our work highlights the uncertainty surrounding the details in the primary
Data Availability Statement: All code used to
sources on the Justinianic Plague and calls into question the likelihood that the Justinianic
complete these simulations and analyses is
publically available on GitHub: https://github.com/ Plague affected all localities in the same way.
whit1951/JustinianPlague. All relevant code is
deposited at Zenodo: https://doi.org/10.5281/
zenodo.3728203.

Funding: This work was supported by NSF DEB

1413925 and the National Socio-Environmental Introduction
Synthesis Center (SESYNC) under funding received
from the NSF DBI 1639145. The funders had no
Since the beginning of the 21st century, scholars have paid increasing attention to the series of
role in study design, data collection and analysis, epidemics that began with the Justinianic Plague, the first part of the earliest plague pandemic
decision to publish, or preparation of the (c. 541–750 CE). Historians have compiled catalogs of the known plague references [1–3], that
manuscript. build upon the ancient primary sources, i.e. texts that were written by contemporaries, who
Competing interests: The authors have declared were often historians (e.g. Prokopios, Evagrios, John of Ephesos). Even so, the limited details
that no competing interests exist. in historical texts have led scholars to question whether the causative agent of Justinianic

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 1 / 21

PLOS ONE Modeling the Justinianic Plague

Plague was truly Yersinia pestis, a debate that was only resolved recently through ancient DNA
analysis [4–7].
Historians have outlined two broad paradigms of the Justinianic Plague. Maximalists
believe that the Justinianic Plague resulted in catastrophic mortality, killing between a quarter
and half of the population of the Eastern Roman Empire, or 15–100 million people, over a few
years [3,5,8]. This narrative often associates plague with the fall of Rome, the end of Antiquity,
and the beginning of the Middle Ages. In contrast, others argue that such estimates of plague
mortality are exaggerated [9–12]. Recent work has pointed to a series of proxy measures—
ranging from contemporary inscriptions to pollen in lakebed sediments—that show no evi-
dence for a major demographic change in the period [13].
Neither side has attempted to employ mathematical models of disease in their arguments.
Instead, maximalist mortality estimates have been based on anecdotal evidence in the primary
sources and simple comparisons with the Black Death. However, unlike the Black Death, the
Justinianic Plague has little concrete evidence. For example, there are no historical documents
for the Justinianic Plague that would allow scholars to reconstruct mortality rates or charts. In
fact, the Justinianic Plague features almost no precise, time-course data that could be used to
validate an epidemiological model of the type that has been used for the Second and Third
Pandemics (e.g. [14–16]).
The complex transmission cycle of plague makes understanding the potential effects of the
Justinianic Plague even more challenging. Modern plague, the closest equivalent to historical
plague [17], relies on numerous sylvatic rodent reservoirs with periodic incursions into
domestic animal and human populations. Bubonic plague in humans may progress to second-
ary pneumonic plague, which can lead to person-to-person transmission that is independent
from the presence of a vector [18,19]. The epidemiological pathway of the Justinianic Plague,
however, remains unclear. Although the main assumption in scholarship is that rats and fleas
are the primary reservoirs and vectors contributing to spillover in humans, there is very little
evidence for enough rats in the premodern Mediterranean to sustain a pandemic (cf. refer-
ences in [12,20,21]). While pneumonic plague and human ectoparasites (i.e. human lice and
fleas), have been suggested as an alternative pathway for the Second Pandemic [16], neither
transmission mode is well supported for the Justinianic Plague and both remain conjectural.
The city of Constantinople in 542 represents the best-documented case study of the first
pandemic by a large margin. A huge city by contemporary standards, Constantinople was the
capital of the Eastern Roman Empire and the permanent place of residence of both the
emperor and the imperial government. The historical sources that report on the outbreak refer
to its temporal shape, length, and a few anecdotal mortality counts. In parallel, historical
research has provided estimates of additional parameters such as the city’s population and gen-
eral life expectancy in antiquity. Since the primary sources provide almost no concrete sup-
porting evidence about the transmission mode of the Justinianic Plague, we developed a series
of dynamic, ordinary differential equation (ODE) models that represent hypothesized trans-
mission routes based on the modern understanding of plague etiology and transmission. We
then compared the output of these models to the anecdotal evidence from the primary sources
to test hypotheses in contemporary historical research about the transmission route and poten-
tial magnitude of this outbreak.
We sought to answer three key questions: (1) given our modern understanding of plague
epidemiology, could a plague outbreak in Constantinople have the impact described by the
historical primary sources, as argued by the maximalist narrative?; (2) which, if any, hypothe-
sized transmission routes would enable an impact of this magnitude?; and (3) within the con-
text of our sensitivity analysis, to what extent are the results from Constantinople generalizable
to other first pandemic outbreaks around the Mediterranean? Although we focus our analysis

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 2 / 21

PLOS ONE Modeling the Justinianic Plague

on the case study of Constantinople in 542 and evaluate the results of our models accordingly,
the uncertainty of the model parameters can serve as a proxy for the variety of Mediterranean
ecosystems affected by the Justinianic Plague.

Methods
Model development
We developed the following ODE models: (1) pneumonic plague (with and without an incuba-
tion period for humans); (2) bubonic plague with the traditional rat, flea, and human dynamics
(with and without an incubation period for humans, rat growth dynamics, and innate resistance
in rats); and (3) bubonic plague that can develop into secondary pneumonic infection and
ongoing pneumonic transmission. We compared time course results using expected parameter
values and compared outcomes from LHS sampling. We evaluated the models based on the rel-
evant primary sources that plague historians use to study the Justinianic Plague.
Initial conditions and parameter estimates. Since historians debate practically all late
antique quantitative historical data, parameter values derived from the historical sources
remain rough estimates. Our parameter values are therefore assumptions based on historical
scholarship. We assumed an initial population size of 500,000 people in Constantinople, based
on the estimates of its population during the outbreak [22]. We also assumed that the human
population size was approximately constant apart from disease-induced mortality, such that
human birth rates were roughly equal to human non-plague related mortality rates and that
mean life expectancy was between 20 and 30 years of age [23].
We used recent literature on plague to derive relevant biological parameter values and
included possible ranges for exploration with Latin Hypercube Sampling (LHS) (Table 1). We
relied on a combination of estimates from empirical infection studies, mechanistic disease
models, and models fitted to historical and present plague outbreaks. Where possible, parame-
ter estimates for hosts and vectors are based on the estimates for the black rat (Rattus rattus)
and the oriental rat flea (Xenopsylla cheopis), which are critical actors in plague transmission
worldwide; both are commonly associated with historical plague [3,24] and frequently appear
in earlier plague studies and models (e.g. [16,25,26]). Expected values were mean values in the
traditional sense for normally distributed parameters, but could also be mode values for asym-
metrical triangle distributions or simply a point estimate for uniform distributions. When
parameter estimates from the literature were highly variable (e.g. estimates of pneumonic
plague transmission rate or transmission rate of bubonic plague from fleas to humans), we
used a uniform sampling distribution to better assess the effects of parameter uncertainty on
model outcomes.
Pneumonic plague (humans only). Pneumonic plague consists of human-to-human air-
borne infection. For pneumonic plague, we considered transmission solely within the human
population and investigated the consequences of demographic rates and an incubation period
on the predicted epidemic dynamics. Without treatment, pneumonic plague has a fatality rate
approaching 100%. Therefore, in accordance with previous pneumonic plague models, we did
not include a recovery class [16,27]. We assumed frequency-dependent transmission since
close contact with an infected individual drives pneumonic plague transmission, requiring
exposure to aerosolized bacteria [18,42]. Beginning with the simplest case of a Susceptible-
Infected-Removed (SIR) compartmental model, the total number of humans (Nh) is given by
the equation:
Nh ðtÞ ¼ Sh ðtÞ þ Ih ðtÞ

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 3 / 21

PLOS ONE Modeling the Justinianic Plague

Table 1. Expected parameter values derived from the literature.

Parameter Description Expected value [Range: min, max] Distribution Reference(s)
rh Human birth rate 1/(25�365) = 0.00011 [1/(30�365) = Uniform [23]; also, consistent with 0.04 yr-1 [26]
0.000091, 1/(20�365) = 0.00013]
days-1
dh Natural human death rate 1/(25�365) = 0.00011 [1/(30�365) = Uniform [23]; also, consistent with 0.04 yr-1 [26]
0.000091, 1/(20�365) = 0.00013]
days-1
βp Pneumonic plague transmission rate in 0.08 [0.01, 1] days-1 Uniform 0.0734 (SE = 0.00005) [27]; 0.08 [28]; 0.084 [19];0.42–0.48
humans [16];
σp−1 Duration of pneumonic incubation rate 4.3 [2.5–6.1] days Normal [27]; 2–4 days [18]
in humans
γp−1 Duration of pneumonic infection 2.5 [1.3–3.7] days Normal [27]
period in humans
βr Transmission rate from fleas to rats 1.248 [0–3.67] fleas-1 days-1 Triangle 0.04–0.14 days-1 [16]; 0–3.67 days-1 � calculated from
percent per flea transmission efficiency/days post
infection at 23˚C [29]
α Flea searching efficiency 3/Sr(t = 0) [0.39 < αKr< 20] rats-1 Uniform [26,30]
rr Reproductive rate in rats 0.014 [0.011, 0.016] rats days-1 Uniform 5/365 days [25,26]
Kr Carrying capacity of rats Nr(t = 0) [0.5Nr(t = 0), Uniform Varied in conjunction with sensitivity analysis on initial
1.5Nr(t = 0)] rats condition (Nr)
dr Natural death rate in rats 0.2 year-1 /365 = 0.00055 [0.1/ Uniform [25,26]
365 = 0.00027, 0.3/365 = 0.00082]
days-1
pr Probability of rats inheriting resistance 0.65 [0.4–0.9] Uniform 0.5 [31]; 0.5–0.9 [32]; 0.4–0.8 [33]
γr−1 Duration of bubonic plague infectious 5.15 [4.71–5.59] days Normal Based on low-moderate infection doses in rats outside of
period in rats plague endemic areas in Madagascar [32]
gr Probability of rats recovering from 0.06 [0.0–0.37] Triangle “24–37% of rats surviving a low- dose infection . . .
bubonic plague compared with only 0–6% at high dose” [32]
rf Growth rate of fleas 0.0084 [0.0084, 20/365 = 0.055] fleas/ Uniform [16,25,26]
day
Kf Flea carrying capacity per rat 6 [3.29, 11.17] fleas Normal [16,25,26]
df−1 Flea lifespan 5 [1, 11.66] days Triangle [16,34]
βb Transmission rate for bubonic plague 0.19 [0.01, 1] days-1 Uniform 0.18–0.20 [16]
from rat fleas to humans
σb−1 Duration of bubonic plague incubation 4 [2, 6] days Triangle [35]; 4.3 days [36]
period in humans
γb−1 Duration of bubonic plague infectious 10 [3, 10] days Triangle 3.4 days [36]; 4–10 days [37]; 10 days [16];
period in humans
gh Probability of humans recovering from 0.34 [0.30, 0.40] Triangle 0.40 [16]; “Among 511 plague cases occurring before
bubonic plague 1942 (pre-antibiotics) with outcome information, 336
(66%) were fatal” [38]
p Probability of human bubonic plague 0.10 [0, 0.15] Triangle 3% [36]; <5% [18]; ~10% [39]; 8–10% [37]; ~12% [40];
developing into secondary pneumonic 5–15% [41]
plague

Here we include the parameter, its description, expected value, range tested during the global sensitivity analysis (LHS-PRCC), and type of distribution used for
sampling.

https://doi.org/10.1371/journal.pone.0231256.t001

The change in the number of susceptible (Sh), infected (Ih), and dead (Dh) humans are
given by the equations:

dSh Ih
¼ bh S h b p Sh dh Sh
dt Nh

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 4 / 21

PLOS ONE Modeling the Justinianic Plague

dIh I
¼ bp Sh h gp Ih
dt Nh

dDh
¼ gp Ih
dt

Extending this to an SEIR framework by adding in an incubation period (σp−1), the total
number of humans becomes: Nh(t) = Sh(t)+Eh(t)+Ih(t)
In addition, the rates of change of the number of susceptible (Sh), exposed (Eh), infected
(Ih), and dead (Dh) humans are given by the equations:

dSh Ih
¼ bh S h b p Sh dh Sh
dt Nh

dEh I dIh
¼ bp Sh h sp Eh ¼ s p Eh gp Ih
dt Nh dt

dDh
¼ gp Ih
dt

Bubonic plague (humans, fleas, and rats). The most common plague model suggested in
the scholarly literature is a transmission cycle maintained through rodents and fleas. Bubonic
plague occurs in wild rodent populations and occasionally spills over to humans through con-
tact with wild or peridomestic rodents [43]. The most common rodent associated with plague
spillover to humans is the black rat (Rattus rattus). Third Pandemic comparatives from 20th
century India have suggested that high mortality among rats precedes human mortality, as
fleas eventually turn to feed upon humans after their preferred rat hosts have died [44].
To model the bubonic plague transmission route, we considered populations of rats (the
primary host), fleas (their vectors), and humans. For rats, dynamics are given in the form of a
Susceptible-Infected-Recovered (SIR) set of equations where the total number of rats (Nr) is
given by:

Nr ðtÞ ¼ Sr ðtÞ þ Ir ðtÞ þ Rr ðtÞ

Ignoring births and deaths in the rat population, the number of susceptible (Sr), infected
(Ir), recovered (Rr), and dead (Dr) rats is given by:

dSr aNr
¼ br Sr F ð1 e Þ=Nr
dt

dIr aNr
¼ br Sr Fð1 e Þ=Nr gr Ir
dt

dRr
¼ gr gr Ir
dt

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 5 / 21

PLOS ONE Modeling the Justinianic Plague

dDr
¼ ð1 gr Þgr Ir
dt

Here βr describes the transmission rate from rats to fleas, α describes the flea searching effi-
ciency, γr−1 describes the bubonic infectious period in rats, and gr describes the probability
that rats survive bubonic plague infection. We assume that the encounter rate between rats
and fleas is represented by a random search process of fleas within a limited area, which is
modulated by the number of available rats (Nr) and the searching efficiency of the fleas (α)
[25,26]. The formulation for this per capita searching efficiency (α) comes from Nicholson &
Bailey’s description of host-parasitoid dynamics [45] and has been used extensively in the
plague modeling literature to describe flea and rat encounter rates [16,25,26]. The model
accounts for the number of expected fleas per rat (H) and the number of free infectious fleas
(F) in the environment that may encounter human hosts.
dH � �
¼ rf H 1 H=Kf
dt

dF
¼ ð1 gr Þgr Ir H df F
dt

Previous modeling studies have identified the sensitivity of this model to the interaction
between carrying capacity (Kr) and flea searching efficiency (α) [25,26]. Recent studies of
endemic plague foci suggest that some rat populations are resistant to plague [31,32]. There-
fore, we also modelled the case where rats grow at a rate modulated by their carrying capacity
(Kr), die at a certain rate (dr), and have a probability (pr) of being born resistant to plague. The
series of equations for the rat population then becomes:
dSr aNr
¼ rr Sr ð1 Nr =Kr Þ þ rr Rr ð1 pr Þ dr Sr br Sr Fð1 e Þ=Nr
dt

dIr aNr
¼ br Sr Fð1 e Þ=Nr gr Ir
dt

dRr
¼ rr Rr ðpr Nr =Kr Þ þ gr gr Ir dr Sr
dt

dDr
¼ ð1 gr Þgr Ir
dt

For the bubonic SIR model, the total number of humans (Nh) is given by: Nh(t) = Sh(t)+Ih(t)
+Rh(t)
In addition, the number of susceptible (Sh), infected (Ih), recovered (Rh), and dead (Dh)
humans are given by the equations:
dSh Sh aNr
¼ bb F ðe Þ þ bh ðSh þ Ih Þ dh Sh
dt Nh

dIh S
¼ b b h F ðe aNr
Þ gb Ih
dt Nh

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 6 / 21

PLOS ONE Modeling the Justinianic Plague

dRh
¼ gh gb Ih dh Rh
dt

dDh
¼ ð1 gh Þgb Ih
dt

For the bubonic SEIR model, including an incubation period, the total number of humans
is represented by the equation: Nh(t) = Sh(t)+Eh(t)+Ih(t)+Rh(t)
The number of susceptible (Sh), exposed (Eh), infected (Ih), recovered (Rh), and dead (Dh)
humans are given by the equations:
dSh Sh aNr
¼ bb F ðe Þ þ bh ðSh þ Ih Þ dh Sh
dt Nh

dEh S
¼ b b h F ðe aNr
Þ sb Eh
dt Nh

dIh
¼ sb Eh gb Ih
dt

dRh
¼ gh gb Ih dh Rh
dt

dDh
¼ ð1 gh Þgb Ih
dt

Bubonic/Pneumonic transmission. In the final scenario, we model a pneumonic plague

outbreak that emerges and is transmitted separately from a bubonic plague transmission cycle.
Humans infected with rat-and-flea transmitted bubonic plague can develop secondary pneu-
monic plague, potentially initiating a chain of human-to-human pneumonic plague transmis-
sion [36,37]. In this case, the total number of humans is described by the equation:

Nh ðtÞ ¼ Sh ðtÞ þ Eb ðtÞ þ Ep ðtÞ þ Ib ðtÞ þ Ip ðtÞ þ Rh ðtÞ

In addition to the susceptible, recovered, and dead classes appearing in prior models, indi-
viduals may experience exposure to bubonic (Eb) or pneumonic plague (Eb) and proceed to
infection with bubonic (Ib) or pneumonic plague (Ip):
dSh Sh aNr Ih
¼ bb F ðe Þ bp Sh þ bh ðSh þ Ih Þ dh Sh
dt Nh Nh

dEb S
¼ bb h Fðe aNr
Þ sb Eh
dt Nh

dEp I
¼ b p Sh h sp Ep
dt Nh

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 7 / 21

PLOS ONE Modeling the Justinianic Plague

dIb
¼ sb Eb gb Ib
dt

dIp
¼ sp Ep þ pgb Ib gp Ip
dt

dRh
¼ gh gb Ib dh Rh
dt

dDh
¼ ð1 p gh Þgb Ih þ gp Ip
dt

Those with bubonic plague may transition to pneumonic plague with a probability, p,
recover with a probability, gh, or die. As outlined in the pneumonic models above, those
infected with pneumonic plague are assumed to have 100% fatality.

Time course results and sensitivity analysis

We solved models numerically using the ode function from the deSolve package in R [46]. For
time course results of bubonic models, we explored initial conditions with different ratios of
rats-to-people: 1:2, 1:1, 2:1 (Fig 1 and S1 and S2 Figs). For all models, we conducted a global
sensitivity analysis using Latin Hypercube Sampling with Partial Ranked Correlation Coeffi-
cients (LHS-PRCC) to evaluate the relative importance of each parameter to the model out-
come by detecting monotonic relationships between parameters and outputs while accounting
for the effects of all other parameters [47]. LHS is a sampling scheme that divides each parame-
ter space into N fractions and samples only once from each of those fractions. Therefore, LHS
is a more efficient sampling method than general Monte Carlo Sampling [48]. The minimum
required sample size (N) for LHS is N�K+1 or N�4/3�K where K is the number of parameters
included in the LHS [49]. The models here range in complexity from 4 to 17 parameters per
model. We created 100 subdivisions per parameter using the lhs package in R (version 3.5.3) to
generate the LHS framework [50].
We conducted PRCC on two outcomes: total mortality and detectable outbreak duration.
We calculated detectable outbreak duration as the number of non-consecutive days where the
mortality rate exceeded 100 deaths per day. To account for multiple comparisons, we used a
Bonferroni altered/corrected p-value (i.e. p = 0.05 divided by the number of model parame-
ters) for calculating confidence intervals. We used 500 bootstraps replicates per sensitivity
index using the pcc function from the sensitivity package when calculating the PRCC values
[51]. All code is deposited at Zenodo: https://doi.org/10.5281/zenodo.3728203.

Model evaluation: Evidence from the primary sources

Although numbers in premodern historical texts are notoriously unreliable, plague maximal-
ists and the majority of broader scholarship have generally accepted the accounts and the num-
bers they provide [4,52]. Since we attempted to test for the feasibility of the maximalist
narrative of the Justinianic Plague, we accepted these numbers at face value for comparison in
our analysis.
The historian John of Ephesus, one eyewitness of the first outbreak, estimated that 300,000
people in Constantinople died during the first outbreak in the city, claiming that officials
stopped counting when they reached 230,000 [53]. Maximalists also cite Prokopios as evidence

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 8 / 21

PLOS ONE Modeling the Justinianic Plague

Fig 1. Time course results of different transmission mode models with rat to human ratio of 1:1. Produced using expected transmission values from
Table 1. Initial conditions: number of susceptible humans, Sh(t = 0) = 500,000, number of susceptible rats, Sr(t = 0) = 499,999, and number of infected rats,
Ir(t = 0) = 1.
https://doi.org/10.1371/journal.pone.0231256.g001

that plague killed half the empire’s population, although the relevant statement is often pre-
sented without context [4,12,54]. Regardless, for the purposes of this paper we estimated the
mortality of the first outbreak in Constantinople as about half the city’s population (i.e.
250,000 people), following the higher end of the range of the maximalist interpretation. Proko-
pios, the most important eyewitness for the Justinianic Plague, supplies additional information
when he asserts that [55]:

“Now the disease in Byzantium [i.e. Constantinople] ran a course of four months, and its
greatest virulence lasted about three. And at first the deaths were a little more than the nor-
mal, then the mortality rose still higher, and afterwards the tale of dead reached five thousand
each day, and again it even came to ten thousand and still more than that.”

As the quote reveals, it is impossible to extract precise numbers from Prokopios’ account.
Historians debate how to treat such vague and suspiciously round estimates, with some choos-
ing to accept these estimates at face value and others remaining more critical towards them
(compare [3,12]). Although the contemporary Roman Empire was a complex bureaucratic
society, how Prokopios or other sources would have known these numbers remains unclear. It

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 9 / 21

PLOS ONE Modeling the Justinianic Plague

seems unlikely that authorities counted corpses when, as the sources say, people stayed off the
streets, corpses remained unburied, and the government shut down. Notably, other ancient
historians refer to reasonable methodologies–for example, after an earthquake in Antioch in
the late sixth century, the historian Evagrios explains his estimate of 60,000 deaths by the num-
ber of people who stopped coming to reclaim their government-supplied free bread [56].
Converting Prokopios’ text into quantitative metrics, we arbitrarily assumed that plague
would be noticeable at about 100 deaths/day above the baseline mortality rate in the city over
all four months, and that the three months with higher mortality should have at least 250 addi-
tional deaths/day. We further use Prokopios’ information about the peak of the mortality rate
during the Constantinople outbreak—namely, his claim that mortality reached 5,000 deaths/
day and eventually 10,000 deaths/day. We chose to follow Prokopios as his relatively detailed
account serves as the basis of modern scholarship on plague and is perceived by scholars to be
a more reliable source that John of Ephesus, who also provides mortality estimates of up to
16,000 deaths/day and the death of 99.9% of the population [53,57].

Results
Model outcomes with expected parameter values
Using expected parameter values, the outbreaks predicted by the pneumonic (SIR and SEIR)
models did not spread successfully and did not exhibit mortality rates exceeding 100 deaths
per day (Fig 1). The results of all five bubonic and bubonic-pneumonic models were relatively
similar and substantially different from the pneumonic models, with estimates of c. 315,000–
340,000 mortalities (63–68% of the city’s population). The two bubonic plague model variants
with rat reproduction and resistance predicted the highest mortality among these models (Fig
1). The mortality in all five models followed a similar trajectory over time. Noticeable mortality
of more than 100 deaths/day began between days 13–15, lasted between 70–76 days, and
ended between days 83–91 (Table 2). The more significant mortality of more than 250 deaths/
day began between days 19–22, continued for between 60–64 days, and ended between days
74–80 (Table 2). The bubonic SEIR model with rat growth and resistance predicted an addi-
tional period of 27 days of high mortality over a month after the previous phase of increased
mortality ended (for a total of 103 days with increased mortality). The bubonic SIR model with
rat growth and resistance predicted a similar trend, albeit below the 100 deaths/day limit
(Table 2).
Outcomes include detectable outbreak duration, maximum mortality rate per day, and total
mortality. Initial conditions: Number of susceptible humans, Sh(t = 0) = 500,000, number of
susceptible rats, Sr(t = 0) = 499,999, and number of infected rats, Ir(t = 0) = 1.

Table 2. Summary of model output for each model type with rat to human ratio of 1:1.
Model Detectable outbreak duration (deaths/day) Maximum mortality rate/day Total mortality (humans)
> 100 > 250
Pneumonic SIR 0 0 0.34 1.25
Pneumonic SEIR 0 0 0.33 1.25
Bubonic SIR 70 60 15152 319170
Bubonic SEIR 74 63 12918 314772
Bubonic SIR (Rat dynamics) 72 61 15116 339989
Bubonic SEIR (Rat 103 (76 consecutive days, 27 additional days after a break of 39 64 12891 339848
dynamics) days)
Bubonic/Pneumonic SEIR 74 64 12778 315489
https://doi.org/10.1371/journal.pone.0231256.t002

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 10 / 21

PLOS ONE Modeling the Justinianic Plague

These findings were relatively robust to changes in the initial conditions of the rat to
human ratio for the bubonic models (S1 and S2 Figs). With a 1:2 ratio of rats to humans, total
mortality decreased to approximately 250,000–257,000 deaths for bubonic SIR, bubonic SEIR,
and bubonic/pneumonic SEIR models (S1 Fig). Bubonic SIR and SEIR models with rat growth
and resistance still predicted roughly 330,000 deaths. However, these outbreaks took substan-
tially longer (374 and 405 days at greater than 100 deaths per day respectively) (S1 Table). A
2:1 ratio of rats to humans increased the convergence of all of the bubonic models and the
bubonic/pneumonic model (S2 Fig). These models predicted human fatalities ranging from
329,000 to 342,000 deaths with 67–71 days exceeding 100 mortalities per day (S2 Table).

Sensitivity analysis: Model outcome variability and parameter influence

Based on the LHS sampling results, the estimated mortality of 250,000 deaths from the primary
sources falls within the interquartile range for all models (Fig 2A). The median outbreak size is
higher for the bubonic SIR and SEIR models with rat dynamics, and the bubonic/pneumonic
SEIR model (Fig 2A). The median outbreak size is lower for the bubonic SIR and SEIR models
and both pneumonic plague models (Fig 2A). Only one of the models contains the detectable
outbreak duration of about 120 days (>100 deaths/day) within their interquartile range:
bubonic SIR with rat dynamics (Fig 2B). For the remaining models, the median outbreak size
and interquartile range is considerably less than the 120 days reported by the primary sources.
None of the models contain the detectable outbreak duration of 90 days (>250 deaths/day)
within their interquartile range. These results were robust to the more conservative approach
of using uniform sampling distributions for all parameters, although the interquartile range of
the detectable outbreak period increased to include the primary source estimates for both
pneumonic plague models (S3 Fig).
For the pneumonic SIR and SEIR models, pneumonic transmission rate (βp) was positively
correlated with outbreak size and detectable duration, and pneumonic plague infectious period
(γp-1) was correlated with decreased outbreak size and detectable duration (S4 Fig). For the
bubonic SIR and SEIR models with and without rat dynamics, flea searching efficiency (α),
flea death rate (df), and rat recovery probability (gr) were generally all negatively correlated
with total mortality (S5 Fig, panels A & C) and duration (S5 Fig, panels B & D). For the
bubonic SIR and SEIR models with rat dynamics, transmission rate from fleas to humans (βb)
also correlated positively with total mortality (S6 Fig).
For the bubonic/pneumonic SIR model, flea searching efficiency (α), longer infectious peri-
ods of pneumonic plague in humans (γp-1), and a shorter pneumonic plague incubation period
(σp-1) were correlated with a higher total human mortality (S7A Fig). Flea searching efficiency
(α) was positively correlated with outbreak duration, while a decreased bubonic incubation
period in humans (σb-1) was correlated with longer detectable outbreaks (S7B Fig). These
parameters generally remained significant with the more conservative sampling approach
using only uniform parameter distributions (S1 Appendix). A full description of LHS PRCC
results and individual correlation plots for each parameter and outcome combination are
available in the S1 and S2 Appendices.

Discussion
Previous retrospective modeling efforts for the Second and Third Pandemics have used precise
outbreak data to statistically estimate unknown transmission parameters (e.g. [27]) or to com-
pare possible transmission routes (e.g. [16]). In the absence of such detailed data for the first
pandemic in Constantinople in 542, we relied on the testimony of primary sources that gave us
rough estimates of total mortality and observed epidemic duration. We then developed a series

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 11 / 21

PLOS ONE Modeling the Justinianic Plague

Fig 2. Box and whisker plot showing results of uniform LHS sampling. (A) number of human mortalities; (B) detectable outbreak duration (>100 deaths per day,
non-consecutive) with inset including outliers (up to 5000 days); and (C) detectable outbreak duration (>250 deaths per day, non-consecutive) with inset including

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 12 / 21

PLOS ONE Modeling the Justinianic Plague

outliers (up to 5000 days). Red lines depict estimated comparison values from historical primary source accounts: (A) 250,000 mortalities; (B) 4 months or 120 days of
mortality exceeding 100 deaths per day; and (C) 3 months or 90 days of mortality exceeding 250 deaths per day.
https://doi.org/10.1371/journal.pone.0231256.g002

of compartmental models to test the possible hypothesized transmission routes for the Justi-
nianic Plague with parameters formed from our modern understanding of plague etiology.

Pneumonic plague
Although the pneumonic plague transmission route remains attractive in some historical liter-
ature because of the higher case mortality associated with it, the primary sources for the Justi-
nianic Plague preserve little evidence for it. Very few sources report symptoms that are
consistent with pneumonic plague. Moreover, since the buboes—that occur in bubonic but
not pneumonic plague—are the most common symptoms reported by far, scholars tend to
diagnose late antique plague based on outbreaks during which the sources refer to buboes. It is
nearly impossible to identify late antique pneumonic plague and differentiate it from any other
epidemic mentioned in the sources.
More broadly, there is little historical evidence for large-scale epidemics of pneumonic
plague: the very high mortality rate of pneumonic plague and its relatively brief incubation
period tend to kill humans too quickly. Moreover, pneumonic plague outbreaks require close
contact between humans and specific environmental conditions. There were two large out-
breaks of pneumonic plague in East Asia in the early 20th century that killed tens of thousands
of people, but all evidence suggests that these required exceptional cultural and environmental
contexts (e.g. [18]).
Using expected parameter values, both our pneumonic plague models (SIR and SEIR) failed
to replicate the historical Justinianic Plague for Constantinople. Neither model predicted suc-
cessful outbreaks (Fig 1), which is not consistent with current maximalist estimates. In addi-
tion, neither model yielded periods where disease-related mortality exceeded 100 deaths per
day, compared to Prokopios’ account that described visible mortality in the city over four
months. Similarly, the highest daily mortality counts in both cases were also considerably
lower than the 10,000 claimed by Prokopios. With LHS sampling, the maximalist estimate of
~250,000 deaths falls within the interquartile range of both the pneumonic SIR and SEIR mod-
els (Fig 2A). However, the model mortalities were largely dichotomous: either failing to foster
onward transmission or decimating the entire population. The median mortalities for both
model formulations were substantially lower than primary source descriptions (Fig 2A), offer-
ing evidence against the idea that pneumonic plague could spread consistently with high mor-
tality through late antique Mediterranean populations.

Bubonic plague
Only limited historical evidence supports the bubonic mode of transmission. Comparisons to
the Third Pandemic suggest that the number of rats required to support a major epidemic
among humans would be very high, as the ratio of infected rats to total rats appears to have
been low in most cases. For example in early 20th century Bombay, perhaps the worst case of
urban plague during the Third Pandemic, less than 3% of 500,000 rats caught had plague [44].
Similarly low ratios of infected rats to total rats commonly appeared in later surveys as well
[58]. A substantial rat population is therefore required to maintain the outbreak among
humans. Rat to human ratios are often estimated at 1:1 in an urban environment [16,44].
However, with the exception of one vague reference in a historical text (to generic “mice” in
a list of animals who died in the plague), none of the surviving late antique accounts of plague

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 13 / 21

PLOS ONE Modeling the Justinianic Plague

mentions rats. This differs significantly from the Third Pandemic evidence that suggests that
preceding rat mortality was very noticeable [44]. There has also been very little archaeological
evidence of late antique rat remains in the many hundreds of archaeological excavations
throughout the Mediterranean [20,59]. The total number of rat bones found in the Eastern
Mediterranean—where we have the most textual evidence for plague—is in the low dozens
over centuries. Rat bones are admittedly quite difficult to find in archaeological excavations,
yet scholars have been searching for them for over a century and have compiled catalogs of
them for over four decades (e.g. [60]).
Although our results for bubonic plague are closer to those of the maximalist interpretation,
they ultimately failed to replicate the historical descriptions of the Justinianic Plague outbreak
in Constantinople. With expected parameter values, the total number of deaths reached
around 315,000–340,000 depending on model construction, which exceeds even the highest
current maximalist estimates for Justinianic Plague mortality (Fig 1). At 15,000 to 19,000, the
maximum deaths per day exceeded Prokopios’ account of 10,000 (Table 2). In contrast, the
total length of increased mortality in most models is shorter than Prokopios’ account with 70–
76 days of mortality exceeding 100 deaths per day (Table 2). Although the SEIR model with rat
growth and resistance resulted in more days of noticeable mortality (103 in total), these days
are split between two periods—a catastrophic period of high mortality, followed by a month-
long lull with a subsequent second month-long increase in observable mortality. This unex-
pected pattern does not appear to conform to the primary source evidence.
For LHS sampling, all bubonic plague model variants contained the mortality estimate of
250,000 deaths within their interquartile ranges (Fig 2A). However, with the exception of the
bubonic SIR model with rat dynamics, none of the models contained the detectable outbreak
duration of 120 days within their interquartile range (Fig 2B). These results similarly do not
conform to the current scholarly understanding of the Justinianic Plague.

Combined bubonic-pneumonic plague

Our third model combined different transmission routes, coupling the higher mortality of
pneumonic plague with the increased persistence of the bubonic rat-flea model, allowing for
continuous bubonic infections that potentially transform into pneumonic plague. With
expected parameter values, the results of this model were consistent with those of the bubonic
models (Fig 1). However, overall mortality reached c. 316,000 (~63% of the population), higher
than the current maximalist estimates (Table 2). The maximum death count per day was c.
12,800, i.e. higher than Prokopios’ estimates, while the length of observable increased mortality
was very close to those of the bubonic model configurations (i.e. 89 days) (Table 2). Looking at
the LHS results, the maximalist estimate of 250,000 deaths is contained within the interquartile
range, but remains lower than the median outcome (Fig 2A). Similarly, the interquartile esti-
mate of detectable outbreak duration is shorter than Prokopios’ account of 120 days (Fig 2B).

Parameter importance and epistemic uncertainty

We conducted a global sensitivity analysis to assess the relative importance of parameters in
dictating model outcomes and to identify epistemic uncertainty for model parameters with
limited empirical backing. Several parameters with high epistemic uncertainty were consis-
tently important across different model configurations. For example, for both the pneumonic
SIR and SEIR models, pneumonic transmission rate (βp) was positively correlated with out-
break size (S4 Fig). Fitted estimates in the literature for contemporary outbreaks range from
0.07–0.46 [16,19,27,28]. In our LHS-PRCC analysis, we conservatively tested a range of 0.01–1
days-1 for this parameter. Our results suggest that when using our proposed structure and

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 14 / 21

PLOS ONE Modeling the Justinianic Plague

parameter values there is a threshold for these pneumonic models where outbreaks are either
completely successful or fail to spread entirely (Fig 2A). Similarly, transmission rate from fleas
to humans (βb) correlated positively with total mortality for both bubonic models with rat
dynamics (S6 Fig), but this parameter has little empirical support for validating a realistic
range.
For bubonic model variants, flea searching efficiency (α) consistently correlated negatively
with total mortality (S5 and S6 Figs). Previous modeling studies have identified the sensitivity
of this model to the interaction between carrying capacity (Kr) and flea searching efficiency (α)
[26,61]. Even so, several modeling studies have treated α as a fixed point value independent of
density (e.g. [30]) or as a fixed value dependent on rat density without acknowledging its con-
tribution to model outcome (e.g. [16]). Our work suggests that a better empirical understand-
ing of the relationship between rat density and flea search efficiency would be important for
future models using this construction to conceptualize encounters between rats and fleas.

Limitations and future directions

Although we tested a variety of model constructions and conducted a comprehensive global
sensitivity analysis, our approach still has several limitations. We investigated multiple hypoth-
esized transmission routes that reflect the modern etiology of Yersinia pestis, but these models
may still not capture the true epidemiology of the Justinianic Plague (cf. [16,25]). Other studies
have highlighted the complexity of the plague transmission cycle. For example, we did not con-
sider the potential of dead hosts (i.e. humans or rats) to serve as reservoirs [30,62], or the myr-
iad of sylvatic hosts contributing to spillover and maintenance of endemic plague [43,63,64].
Similarly, consistent with other modeling studies (e.g. [16,25,26]), we did not differentiate
between early phase transmission and blocked transmission, wherein Y. pestis produces a bio-
film blocking the flea foregut, both increasing transmission efficiency and decreasing flea life
span [30,62]. Finally, flea life span and Y. pestis transmission efficiency is temperature depen-
dent [29,34]; we did not explore the consequences of seasonal variation on epidemic outcomes
(e.g. [65]). A related issue is that the genetic composition of the Y. pestis that is associated with
the Justinianic Plague is slightly different from that of the current day Y. pestis [7,17]. We
therefore cannot rule out the possibility that the late antique Y. pestis was more or less lethal to
humans.
Other factors that we cannot completely dismiss include possible differences between late
antique and modern hygiene, public health practices, and reduced human immunocompe-
tence due to increased risk of comorbidities or poor nutritional status. Discussions in the his-
torical literature have just begun examining the possibility of nutritional stress and
coinfection, although data remains anecdotal and incomplete (e.g. [3,66]) We have no data
about where in Constantinople plague-related mortality took place, or about specific practices
and locations of waste disposal in the city that might be associated with higher rat populations.
Although our conservative parameter estimates (Table 1) and LHS approach may capture
some of these dimensions, we did not attempt to explicitly model differences in hygiene or
comorbidity with different diseases as any such analysis would be conjectural and based on
very sparse data at best.
For human birth and natural death rates, we assumed average life expectancy of 25 years
(range 20–30 years). In reality, however, non-plague related mortality in this period would
have been much higher in infancy and pre-adolescence with the likelihood of surviving to an
older age (50–60 years and beyond) improving after these earlier stages of life ([22] citing
[67]). Future extensions of this model could relax the assumption of exponentially distributed
demographic rates [68]. Our models did not consider any social response to plague, although

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 15 / 21

PLOS ONE Modeling the Justinianic Plague

contemporaries could have mitigated their risk through several simple strategies such as leav-
ing the city or minimizing social contact. The limited data available to us led us to model the
most important city within the empire; although the results could, with some reservations, be
suggestive of other urban centers in the empire, they would say nothing about the empire’s
rural population, which made up the vast majority of its total population.
It is possible that the primary sources that we used to evaluate our models are themselves
exaggerated, and that the demographic effects of the Justinianic Plague were far lower than
expected. As suggested above, this interpretation is consistent with the general skepticism his-
torians have with regard to premodern primary sources, and with the fact that scholarly esti-
mates of parameter values vary by orders of magnitude even when describing fairly concrete
values such as the number of deaths. This line of reasoning has been suggested before, consid-
ering the general absence of non-textual evidence for the Justinianic Plague causing extremely
high mortality in Constantinople and beyond. While this paper does not prove that these pri-
mary sources and maximalist studies are wrong, it casts doubt on such maximalist
interpretations.
Although the model results did not corroborate the few details that the primary sources
supply, the variability of these model results is potentially important for our broader under-
standing of the Justinianic Plague and its effects across Mediterranean urban centers. Assum-
ing that the variability in the models’ parameters would reflect variability in the environmental
(e.g. temperature, precipitation, humidity), ecological (e.g. rat and flea densities) and social
systems (e.g. contact patterns) around the Mediterranean, the spread of the results of all mod-
els suggests that plague’s impacts on other urban areas could differ substantially from its
impacts on Constantinople. This in turn confirms the fragmentary evidence for the First Pan-
demic, where certain areas seem to be associated with plague more often than others, and
empirical evidence from the Third Pandemic that indicates that plague mortality varied con-
siderably both temporally and spatially (e.g. table in [69]).
Nonetheless, since we evaluated these models in a deterministic framework, we cannot rule
out the effects of stochasticity that could lead to probabilistically low, but singularly large out-
breaks [27]. Future modeling extensions could extend this framework to allow for stochasticity
and more realistic spatial structure. Our models assumed that all the city’s inhabitants mixed
homogeneously within the same population; additional realism could be added by incorporat-
ing a metapopulation structure to reflect the uneven contact mixing structure of a large urban
metropolis (e.g. [25,26]) or by using a spatial network modeling framework to examine how
travel and trading routes affect the potential for plague outbreaks [70]).

Conclusions
In this paper, we have attempted to reconstruct the common scenario of the Justinianic Plague
that former literature accepts. According to this scenario, between a quarter and half the popu-
lation of Constantinople perished during the first outbreak of plague in the city in 542. Despite
their general agreement on this conclusion, historians have proposed different models to
explain the plague epidemiology that appears in the historical texts that serve as the foundation
of modern plague narratives. Instead of selecting one of these epidemiological models, we
tested them all with parameter values supported by our modern understanding of plague etiol-
ogy. While several of the bubonic plague model configurations yielded interquartile estimates
arguably consistent with the upper end of maximalist estimates of mortality, plague in these
models lasted for considerably less time than suggested by the primary sources. By contrast,
both pneumonic models suggest that pneumonic plague by itself was unlikely to cause almost
any mortality in Constantinople. When viewed in light of the historical evidence, the model

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 16 / 21

PLOS ONE Modeling the Justinianic Plague

results therefore indicate that the outbreak in Constantinople was very likely not a pneumonic
plague outbreak.
Our results suggest that given what we know of modern plague etiology it would have been
highly unlikely for a plague outbreak to have the magnitude of impact with the simultaneous
outbreak duration that the primary sources describe. Since the outbreak at Constantinople
underlies the scholarly understanding of the broader Justinianic Plague and has the most evi-
dence compared to outbreaks elsewhere or later in the first pandemic, our results suggest that
the Justinianic Plague behaved differently than the current maximalist consensus postulates,
and thus contribute to the broader discussion of the impact of plague during Late Antiquity.

Supporting information
S1 Fig. Time course results of different transmission mode models with rat to human ratio
of 1:2. Produced using expected transmission values from Table 1. Initial conditions: number
of susceptible humans, Sh(t = 0) = 500,000, number of susceptible rats, Sr(t = 0) = 249,999, and
number of infected rats, Ir(t = 0) = 1.
(TIF)
S2 Fig. Time course results of different transmission mode models with rat to human ratio
of 2:1. Produced using expected transmission values from Table 1. Initial conditions: number
of susceptible humans, Sh(t = 0) = 500,000, number of susceptible rats, Sr(t = 0) = 999,999, and
number of infected rats, Ir(t = 0) = 1.
(TIF)
S3 Fig. Box and whisker plot showing results of uniform LHS sampling. (A) number of
human mortalities; and (B) detectable outbreak duration (>100 deaths per day, nonconsecu-
tive) with inset including outliers (up to 5000 days); and (C) detectable outbreak duration
(>250 days, nonconsecutive) with inset including outliers (up to 5000 days) zooming in on
time axis of 0 to 1000 days. Red lines depict estimated comparison values from historical pri-
mary source accounts: (A) 250,000 mortalities; (B) 4 months or 120 days of mortality exceed-
ing 100 deaths per day; and (C) 3 months or 90 days of mortality exceeding 250 deaths per
day.
(TIF)
S4 Fig. LHS-PRCC results for pneumonic plague SIR and SEIR models. (A) Total mortality
for SIR; (B) detectable outbreak duration (days) for SIR; (C) total mortality for SEIR; and (D)
detectable outbreak duration (days) for SEIR.
(TIF)
S5 Fig. LHS-PRCC results for bubonic plague SIR and SEIR models. (A) Total mortality for
SIR; (B) detectable outbreak duration (days) for SIR; (C) total mortality for SEIR; and (D)
detectable outbreak duration (days) for SEIR.
(TIF)
S6 Fig. LHS-PRCC results for bubonic plague SIR and SEIR models with rat dynamics. (A)
Total mortality for SIR; (B) detectable outbreak duration (days) for SIR; (C) total mortality for
SEIR; and (D) detectable outbreak duration (days) for SEIR.
(TIF)
S7 Fig. LHS-PRCC results for bubonic/pneumonic plague SEIR model. (A) Total mortality
for SIR and (B) detectable outbreak duration (days).
(TIF)

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 17 / 21

PLOS ONE Modeling the Justinianic Plague

S1 Table. Summary of model output for each model type with rat to human ratio of 1:2.
Outcomes include detectable outbreak duration, maximum mortality rate per day, and total
mortality. Initial conditions: number of susceptible humans, Sh(t = 0) = 500,000, number of
susceptible rats, Sr(t = 0) = 249,999, and number of infected rats, Ir(t = 0) = 1.
(DOCX)
S2 Table. Summary of model output for each model type with rat to human ratio of 2:1.
Outcomes include detectable outbreak duration, maximum mortality rate per day, and total
mortality. Initial conditions: number of susceptible humans, Sh(t = 0) = 500,000, number of
susceptible rats, Sr(t = 0) = 999,999, and number of infected rats, Ir(t = 0) = 1.
(DOCX)
S1 Appendix. LHS uniform sampling results. Includes individual scatter plots of parameters
vs. model outcomes and PRCC plots.
(DOCX)
S2 Appendix. LHS non-uniform sampling results. Includes individual scatter plots of param-
eters vs. model outcomes and PRCC plots.
(DOCX)

References
1. Biraben J-N, Le Goff J. The plague in the early Middle Ages. In: Forster R, Ranum O, editors. Biology of
man in history. Baltimore: The Johns Hopkins University Press; 1975. pp. 48–80.
2. Stathakopoulos DC. Famine and pestilence in the late Roman and early Byzantine Empire: a systematic
survey of subsistence crises and epidemics. Routledge; 2017. Available: https://play.google.com/
store/books/details?id=KgwkDwAAQBAJLB.pf9D
3. Harper K. The fate of Rome. Princeton: Princeton University Press; 2017. https://doi.org/10.1515/
9781400888917
4. Harbeck M, Seifert L, Hänsch S, Wagner DM, Birdsell D, Parise KL, et al. Yersinia pestis DNA from skel-
etal remains from the 6(th) century AD reveals insights into Justinianic Plague. PLoS Pathog. 2013; 9:
e1003349. https://doi.org/10.1371/journal.ppat.1003349 PMID: 23658525
5. Wagner DM, Klunk J, Harbeck M, Devault A, Waglechner N, Sahl JW, et al. Yersinia pestis and the
plague of Justinian 541–543 AD: A genomic analysis. Lancet Infect Dis. 2014; 3099: 1–8. https://doi.
org/10.1016/S1473-3099(13)70323-2
6. Feldman M, Harbeck M, Keller M, Spyrou MA, Rott A, Trautmann B, et al. A high-coverage Yersinia
pestis genome from a sixth-century Justinianic Plague victim. Mol Biol Evol. 2016; 33: 2911–2923.
https://doi.org/10.1093/molbev/msw170 PMID: 27578768
7. Keller M, Spyrou MA, Scheib CL, Neumann GU, Kröpelin A, Haas-Gebhard B, et al. Ancient Yersinia
pestis genomes from across Western Europe reveal early diversification during the First Pandemic
(541–750). Proc Natl Acad Sci U S A. 2019; 116: 12363–12372. https://doi.org/10.1073/pnas.
1820447116 PMID: 31164419
8. Meier M. The ‘Justinianic Plague’: The economic consequences of the pandemic in the eastern Roman
empire and its cultural and religious effects. Early Mediev Eur. 2016; 24: 267–292. https://doi.org/10.
1111/emed.12152
9. Durliat J. La peste du VIe siècle: Pour un nouvel examen des sources Byzantines. In: Abadie-Reynal C,
Morrisson C, Lefort J, editors. Hommes et richesses dans l’Empire byzantin. Lethielleux; 1989. pp.
107–119.
10. Wickham C. Framing the early Middle Ages. Oxford University Press; 2005. https://doi.org/10.1093/
acprof:oso/9780199264490.001.0001
11. Whittow M. The making of Orthodox Byzantium, 600–1025. London: Red Globe Press; 1996. https://
doi.org/10.1007/978-1-349-24765-3
12. Mordechai L, Eisenberg M. Rejecting catastrophe: The case of the Justinianic Plague*. Past Present.
2019; 244: 3–50. https://doi.org/10.1093/pastj/gtz009
13. Mordechai L, Eisenberg M, Newfield TP, Izdebski A, Kay JE, Poinar H. The Justinianic Plague: An
inconsequential pandemic? Proc Natl Acad Sci U S A. 2019; 116: 25546–25554. https://doi.org/10.
1073/pnas.1903797116 PMID: 31792176

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 18 / 21

PLOS ONE Modeling the Justinianic Plague

14. Whittles LK, Didelot X. Epidemiological analysis of the Eyam plague outbreak of 1665–1666. Proc R
Soc B Biol Sci. 2016; 283: 20160618. https://doi.org/10.1098/rspb.2016.0618 PMID: 27170724
15. Dean KR, Schmid B V., Krauer F, Schmid B V. Epidemiology of a bubonic plague outbreak in Glasgow,
Scotland in 1900. R Soc Open Sci. 2019; 6: 181695. https://doi.org/10.1098/rsos.181695 PMID:
30800398
16. Dean KR, Krauer F, Walløe L, Lingjærde OC, Bramanti B, Stenseth NC, et al. Human ectoparasites
and the spread of plague in Europe during the Second Pandemic. Proc Natl Acad Sci. 2018; 115:
201715640. https://doi.org/10.1073/pnas.1715640115 PMID: 29339508
17. Zhou Z, Alikhan N-F, Mohamed K, Group the AS, Achtman M. The user’s guide to comparative geno-
mics with EnteroBase. Three case studies: Micro-clades within Salmonella enterica serovar Agama,
ancient and modern populations of Yersinia pestis, and core genomic diversity of all Escherichia. bioR-
xiv. 2019. https://doi.org/10.1101/613554
18. Kool JL, Weinstein RA. Risk of person-to-person transmission of pneumonic plague. Clin Infect Dis.
2005; 40: 1166–1172. https://doi.org/10.1086/428617 PMID: 15791518
19. Ratsitorahina M, Chanteau S, Rahalison L, Ratsifasoamanana L, Boisier P. Epidemiological and diag-
nostic aspects of the outbreak of pneumonic plague in Madagascar. Lancet. 2000; 355: 111–113.
https://doi.org/10.1016/S0140-6736(99)05163-6 PMID: 10675169
20. McCormick M. Rats, communications, and plague: Toward an ecological history. J Interdiscip Hist.
2003; 34: 1–25. https://doi.org/10.1162/002219503322645439
21. Hufthammer AK, Walløe L. Rats cannot have been intermediate hosts for Yersinia pestis during medie-
val plague epidemics in Northern Europe. J Archaeol Sci. 2013; 40: 1752–1759. https://doi.org/10.
1016/j.jas.2012.12.007
22. Croke B. Justinian’s Constantinople. In: Mass M, editor. The Cambridge companion to the Age of Jus-
tinian. Cambridge: Cambridge University Press; 2005. pp. 60–86. https://doi.org/10.1017/
CCOL0521817463.021
23. Frier BW. Demography. In: Bowman A, Garnsey P, Rathbone D, editors. The Cambridge ancient history
XI: The high empire. Cambridge: Cambridge University Press; 2000. pp. 787–816. https://doi.org/10.
1017/CHOL9780521263351.028
24. Horden P. Mediterranean plague in the Age of Justinian. In: Mass M, editor. The Cambridge companion
to the Age of Justinian. Cambridge University Press LB—WRE6; 2005. pp. 134–160.
25. Keeling MJ, Gilligan CA. Metapopulation dynamics of bubonic plague. Nature. 2000; 407: 903–6.
https://doi.org/10.1038/35038073 PMID: 11057668
26. Keeling MJ, Gilligan CA. Bubonic plague: a metapopulation model of a zoonosis. Proc R Soc London B
Biol Sci. 2000; 267: 2219–2230. https://doi.org/10.1098/rspb.2000.1272 PMID: 11413636
27. Gani R, Leach S. Epidemiologic determinants for modeling pneumonic plague outbreaks. Emerg Infect
Dis. 2004; 10: 608–14. https://doi.org/10.3201/eid1004.030509 PMID: 15200849
28. Begier EM, Asiki G, Anywaine Z, Yockey B, Schriefer ME, Aleti P, et al. Pneumonic plague cluster,
Uganda, 2004. Emerg Infect Dis. 2006; 12: 460–467. https://doi.org/10.3201/eid1203.051051 PMID:
16704785
29. Schotthoefer AM, Bearden SW, Holmes JL, Vetter SM, Montenieri JA, Williams SK, et al. Effects of tem-
perature on the transmission of Yersinia pestis by the flea, Xenopsylla cheopis, in the late phase period.
Parasit Vectors. 2011; 4: 1–11. https://doi.org/10.1186/1756-3305-4-1 PMID: 21205315
30. Webb CT, Brooks CP, Gage KL, Antolin MF. Classic flea-borne transmission does not drive plague epi-
zootics in prairie dogs. Proc Natl Acad Sci. 2006; 103: 6236–6241. https://doi.org/10.1073/pnas.
0510090103 PMID: 16603630
31. Gascuel F, Choisy M, Duplantier J-M, Débarre F, Brouat C. Host resistance, population structure and
the long-term persistence of bubonic plague: Contributions of a modelling approach in the Malagasy
focus. PLOS Comput Biol. 2013; 9: e1003039. https://doi.org/10.1371/journal.pcbi.1003039 PMID:
23675291
32. Tollenaere C, Rahalison L, Ranjalahy M, Duplantier J-MM, Rahelinirina S, Telfer S, et al. Susceptibility
to Yersinia pestis experimental infection in wild Rattus rattus, reservoir of plague in Madagascar. Eco-
health. 2010; 7: 242–247. https://doi.org/10.1007/s10393-010-0312-3 PMID: 20443044
33. Biggins DE, Kosoy MY. Influences of introduced plague on North American mammals: Implications
from ecology of plague in Asia. J Mammology. 2001; 82: 906–916. https://doi.org/10.1644/1545-1542
(2001)082<0906:ioipon>2.0.co;2
34. Bacot AW, Martin CJ. The respective influences of temperature and moisture upon the survival of the
rat flea (Xenopsylla cheopis) away from its host. J Hyg. 1924; 23: 98–105. https://doi.org/10.1017/
s0022172400008500 PMID: 20474832

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 19 / 21

PLOS ONE Modeling the Justinianic Plague

35. Frequently Asked Questions | Plague | CDC. 2018. Available: https://www.cdc.gov/plague/faq/index.

htmlLB—pLww
36. Nguyen VK, Parra-Rojas C, Hernandez-Vargas EA. The 2017 plague outbreak in Madagascar: Data
descriptions and epidemic modelling. Epidemics. 2018; 25: 20–25. https://doi.org/10.1016/j.epidem.
2018.05.001 PMID: 29866421
37. Nikiforov V V, Gao H, Zhou L, Anisimov A. Plague: Clinics, diagnosis and treatment. Adv Exp Med Biol.
2016. pp. 293–312. https://doi.org/10.1007/978-94-024-0890-4_11 PMID: 27722868
38. Kugeler KJ, Erin Staples J, Hinckley AF, Gage KL, Mead PS, Staples JE, et al. Epidemiology of human
plague in the United States, 1900–2012. Emerg Infect Dis. 2015; 21: 16–22. https://doi.org/10.3201/
eid2101.140564 PMID: 25529546
39. Alsofrom DJ, Mettler FA, Mann JM. Radiographic manifestations of plaque in New Mexico, 1975–1980.
A review of 42 proved cases. Radiology. 1981; 139: 561–5. https://doi.org/10.1148/radiology.139.3.
7232721 PMID: 7232721
40. Inglesby T V, Dennis DT, Henderson DA, Bartlett JG, Ascher MS, Eitzen E, et al. Plague as a biological
weapon. JAMA. 2000; 283: 2281. https://doi.org/10.1001/jama.283.17.2281 PMID: 10807389
41. Darling RG, Catlett CL, Huebner KD, Jarrett DG. Threats in bioterrorism. I: CDC category A agents.
Emerg Med Clin North Am. 2002; 20: 273–309. Available: https://www.ncbi.nlm.nih.gov/pubmed/
12120480 https://doi.org/10.1016/s0733-8627(02)00005-6 PMID: 12120480
42. Keeling M, Rohani P. Modeling infectious diseases in humans and animals. Princeton, New Jersey:
Princeton University Press; 2008.
43. Gage KL, Kosoy MY. Natural history of plague: perspectives from more than a century of research.
Annu Rev Entomol. 2005; 50: 505–28. https://doi.org/10.1146/annurev.ento.50.071803.130337 PMID:
15471529
44. Turner JA. Sanitation in India. The Times of India; 1914.
45. May RM, Hassell MP, Anderson RM, Tonkyn DW. Density dependence in host-parasitoid models. J
Anim Ecol. 1981; 50: 855. https://doi.org/10.2307/4142
46. Setzer SKP. Solving differential equations in R: Package deSolve. J Stat Softw. 2010; 33: 1–25. https://
doi.org/10.18637/jss.v033.i09 PMID: 20808728
47. Marino S, Hogue IB, Ray CJ, Kirschner DE. A methodology for performing global uncertainty and sensi-
tivity analysis in systems biology. J Theor Biol. 2008; 254: 178–196. https://doi.org/10.1016/j.jtbi.2008.
04.011 PMID: 18572196
48. Wu J, Dhingra R, Gambhir M, Remais J V. Sensitivity analysis of infectious disease models: Methods,
advances and their application. J R Soc Interface. 2013; 10: 20121018. https://doi.org/10.1098/rsif.
2012.1018 PMID: 23864497
49. Blower SM, Dowlatabadi H. Sensitivity and uncertainty analysis of complex models of disease transmis-
sion: An HIV model, as an example. Int Stat Rev. 1994; 62: 229–243. https://doi.org/10.2307/1403510
50. Carnell R. lhs: Latin hypercube samples. 2019. Available: https://cran.r-project.org/package=lhsLB—
kVLh
51. Iooss B, Janon A, Pujol G, Broto B, Boumhaout K, Da Veiga S, et al. sensitivity: Global sensitivity analy-
sis of model outputs. 2019. Available: https://cran.r-project.org/package=sensitivity
52. Mitchell S. A history of the later Roman Empire, AD 284–641. John Wiley & Sons; 2014.
53. Ps. Dionysius of Tel-Mahre. The chronicle of Zuqnīn parts III and IV A.D. 488–775. Pontifical Institute
of Mediaeval Studies; 1999.
54. Prokopios. Secret history. Dewing H, editor. The Macmillan Co.; 1914.
55. Prokopios. Wars. Dewing H, editor. The Macmillan Co.; 1914.
56. Evagrios (Scholasticus), Whitby M. Evagrios 6.8. The ecclesiastical history of Evagrius Scholasticus.
Liverpool University Press; 2000.
57. Michael the Syrian. Chronique de Michel le Syrien: patriarche jacobite d’Antioche (1166–1199). Ernest
Leroux; 1899.
58. Bramanti B, Dean KR, Walløe L, Chr Stenseth N. The third plague pandemic in Europe. Proc Biol Sci.
2019; 286: 20182429. https://doi.org/10.1098/rspb.2018.2429 PMID: 30991930
59. Audoin-Rouzeau F, Vigne J-D. La colonisation de l’Europe par le rat noir (Rattus rattus). Rev Paléobio-
logie. 1994; 13: 125–145.
60. Becker K. Rattus rattus. In: Niethammer J, Krapp F, editors. Handbuch der Säugetiere Europas, Vol 1.
Akademische Verlagsgesellschaft; 1978. pp. 382–400.

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 20 / 21

PLOS ONE Modeling the Justinianic Plague

61. Buzby M, Neckels D, Antolin MF, Estep D. Analysis of the sensitivity properties of a model of vector-
borne bubonic plague. J R Soc Interface. 2008; 5: 1099–1107. https://doi.org/10.1098/rsif.2007.1339
PMID: 18270149
62. Gandon S, Heitzmann L, Sebbane F. To block or not to block: The adaptive manipulation of plague
transmission. Evol Lett. 2019; 3: 152–161. https://doi.org/10.1002/evl3.111 PMID: 31161047
63. Wimsatt J, Biggins DE. A review of plague persistence with special emphasis on fleas. J Vector Borne
Dis. 2009; 46: 85–99. PMID: 19502688
64. Foley JE, Zipser J, Chomel B, Girvetz E, Foley P. Modeling plague persistence in host-vector communi-
ties in California. J Wildl Dis. 2007; 43: 408–424. https://doi.org/10.7589/0090-3558-43.3.408 PMID:
17699079
65. Lewnard JA, Townsend JP. Climatic and evolutionary drivers of phase shifts in the plague epidemics of
colonial India. Proc Natl Acad Sci. 2016; 113: 14601–14608. https://doi.org/10.1073/pnas.1604985113
PMID: 27791071
66. Newfield TP. Mysterious and mortiferous clouds: The climate cooling and disease burden of Late Antiq-
uity. Late Antiq Archaeol. 2018; 12: 89–115. https://doi.org/10.1163/22134522-12340068
67. Coale AJ, Demeny P, Vaughan B. Regional model life tables and stable Populations. 2nd ed. Aca-
demic Press; 1984. https://doi.org/10.2307/2067597 LB
68. Feng Z, Xu D, Zhao H. Epidemiological models with non-exponentially distributed disease stages and
applications to disease control. Bull Math Biol. 2007; 69: 1511–1536. https://doi.org/10.1007/s11538-
006-9174-9 PMID: 17237913
69. Gait EA. Census of India, 1911. Vol. I: India. Part I: Report. Superintendent Government Printing,
India; 1913.
70. Gómez JM, Verdú M. Network theory may explain the vulnerability of medieval human settlements to
the Black Death pandemic. Sci Rep. 2017; 7: 43467. https://doi.org/10.1038/srep43467 PMID:
28262733

PLOS ONE | https://doi.org/10.1371/journal.pone.0231256 April 30, 2020 21 / 21