www.nature.

com/scientificreports

Corrected: Author Correction

OPEN Antibiotic drug-resistance as a

complex system driven by socio-
economic growth and antibiotic
Received: 19 February 2019
Accepted: 17 June 2019 misuse
Published online: 05 July 2019
Bhawna Malik & Samit Bhattacharyya

Overwhelming antibiotic use poses a serious challenge today to the public-health policymakers
worldwide. Many empirical studies pointed out this ever-increasing antibiotic consumption as primary
driver of the community-acquired antibiotic drug-resistance, especially in the middle- and lower-
income countries. The association is well documented across spatio-temporal gradients in many parts
of the world, but there is rarely any study that emphasizes the mechanism of the association, which
is important for combating drug-resistance. Formulating a mathematical model of emergence and
transmission of drug-resistance, we in this paper, present how amalgamating three components: socio-
economic growth, population ecology of infectious disease, and antibiotic misuse can instinctively
incite proliferation of resistance in the society. We show that combined impact of economy, infections,
and self-medication yield synergistic interactions through feedbacks on each other, presenting the
emergence of drug-resistance as a self-reinforcing cycle in the population. Analysis of our model not
only determines the threshold of antibiotic use beyond which the emergence of resistance may occur,
but also characterizes how fast it develops depending on economic growth, and lack of education
and awareness of the population. Our model illustrates that proper and timely government aid in
population health can break the self-reinforcing process and reduce the burden of drug-resistance in the
community.

The rapid emergence and dissemination of resistant bacteria is occurring globally, threatening the potency of
antibiotics, which saves millions of lives1,2. It is described as the primary threat to public health in the 21st cen-
tury, and agencies around the globe have underlined the urgent need for actions3,4. There is high resistance to
Cotrimoxazole (68.5%) and Ampicillin (84.5%) among the Gram-negative bacilli in Zimbabwe5. In the same
country, Gram-positive cocci is also resistant to Cotrimoxazole (69%) and Nalidixic acid (81%). E. coli is another
pathogen that shows 84% resistance to ampicillin6. A recent report has documented the non-susceptibility
pattern of key Gram-negative (such as Klesbsiella, E. coli, Salmonella, Shigella) and Gram-positive pathogens
(such as S. pneumoniae, S. aureus, S. agalactae, MRSA) to the most common antibiotics in Sub-Saharan Africa7.
However, antibiotic resistance is worldwide, and the effects of resistance are more severe in lower- and upper
middle-income countries (Fig. S1)8,9. This ever-growing problem not only threatens public health, but also incurs
a huge toll on nation’s economic growth by delayed hospitalizations, lengthening recovery time, expensive med-
icines, and specialized care for patients10–12. Shrestha et al. (2018) have estimated the economic cost of resistance
per antibiotic by drug class and compared those in developing and developed countries, like in Thailand and
United States11. Moreover, mortality and morbidity are the indirect consequences of resistance that incur a huge
loss in income leading the sufferer and family to the poor economic condition13.
Antibiotic resistance may occur naturally, but irrational use of antibiotics accelerate the process leading to
expulsion of sensitive strain, and procreation and propagation of the resistant bacteria in the community11,14,15.
Several countries already have taken steps to reduce the misuse of antimicrobials. World Health Organization
(WHO) already has launched a Global Action Plan in 2015 to initiate evidence based prescribing through effec-
tive, rapid, and low-cost diagnostic tools to optimize use of antimicrobials16. In spite of these efforts, the volume

Disease Modelling Lab, Department of Mathematics, School of Natural Sciences, Shiv Nadar University, Gautan Buddha
Nagar, India. Correspondence and requests for materials should be addressed to B.M. (email: [email protected]) or
S.B. (email: [email protected])

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 1

www.nature.com/scientificreports/ www.nature.com/scientificreports

Population ecology
of infectious
diseases

Antibiotic
drug-resistance as
complex system

Antibiotic Socio-economic
Self-medication /
consumption growth
Over-the-counter (OTC)

Figure 1.  Antibiotic drug-resistance as an integrated system driven by population ecology of infectious disease,
socio-economic growth of population and antibiotic consumption by individuals in the population.

of antibiotic use is ever-increasing worldwide, especially in developing countries. A recent report indicates that
the global antibiotic consumption increased by 65% (rate increased by 39%) between 2000 and 2015, from 21.1
to 34.8 billion Daily Defined Doses (DDDs), and the global increase was primarily driven by increased consump-
tion in Lower Middle Income Countries (Fig. S2)9. Countries like Brazil, Russia, India, China, and South Africa
(BRICS countries) are developing nations, which show the highest drug consumption from 2000 to 2010 with
India as the first and China is in the second position17.
Self-medication plays a major role in ever-increasing antibiotic consumption18–21. Over-the-counter (OTC)
sales – which is more ubiquitous in more economically destitute society – is one of the source of self-medication22.
It is a very common practice in countries like Bangladesh23, Addis Ababa and Central Ethiopia24, and North West
Ethiopia25. Avoiding expenditure of treatment, inability to access medical facilities, less education and lack of
awareness are utmost challenges for self-medication26. Also, perceived patient expectation, lack of knowledge
and diagnostics, incentives and advertising from industry, and financial benefits are major factors for irrational
antibiotic prescription by physicians in most of the developing countries27.
The relation between antimicrobial use and emergence of resistance is a complex association. While most of
the recent research focus on understanding microbiology of drug-resistance, there are only few modelling studies
that emphasize transmission dynamics and coexistence of strains under volume of antibiotic use. For example,
Massad et al. (1993) studied coexistence of sensitive and resistant strains in hospital infection28. Austin et al.
(1999, 2001) developed model to understand influence of drug use on transmission of resistance29. In the past
few years, various antibiotic drug-resistance models have been developed that discuss usage of antibiotics and
emergence of resistance30–34. For example, Levin et al. (2014) have shown by developing simple epidemiological
model that burden of resistance can be maintained under certain acceptable level if the use of specific antibiot-
ics declines with the frequency of resistance to these drugs35. Developing game theoretic model, Fu and Chen
(2018) have shown how social learning may help prescribing behavior of physicians to promote social optimum
of antibiotic consumption36. However, there is rarely any modelling work that integrate socio-economic growth,
antibiotic use and transmission dynamics to understand this ever-increasing antibiotic consumptions in middle
and lower income countries.
In this paper, we present a mechanistic model of community-acquired drug-resistance, combining population
ecology of infectious disease, economic growth and antibiotic use as a function of individual economic status.
Every country has its own economic constraints and epidemiological parameters behind the emergence and prev-
alence of drug resistance in its population. Also, different pathogens have different mechanisms of emergence and
transmission. Instead of incorporating specific characteristics of a diverse variety of pathogens, we test our theory
using general susceptible-infected-susceptible (SIS) model, where individuals can have recurrent infections over
the course of their lifetime. This kind of general framework of reinfection is considered to represent the repeated
threat of infection that individuals face in most of developing tropical countries. However, our model can demon-
strate the emergence of antibiotic drug-resistance in the population as a self-reinforcing system where these three
components interact through positive feedbacks on each other (Fig. 1). It can explain the ever-increasing develop-
ment of resistance, especially in lower-income countries. We interpret how invasion and dominance of resistance
are correlated with socio-economic growth and antibiotic consumption in the population. Our model analyses
reveal that a large inflow of capital in the form of development aid, especially in the early stage, can escape the
situation by encouraging economic growth and scaling down the antibiotic misuse.

Material and Methods

Our drug-resistance model connects three domains of human population: socio-economic growth, ecology of infec-
tious disease, and antibiotic (mis)use by individuals. Although these three aspects are dynamic in nature and
canonical in their respective disciplines, combining them may explain the evolution of drug-resistance in the
population. Below, we describe model formulation in these three parts separately, and then develop the integrated
model. Detailed model development of separate components are discussed in the Supplementary Information.

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 2

www.nature.com/scientificreports/ www.nature.com/scientificreports

Figure 2.  Schematic of the model. For detail explanation, see the Supplementary Information.

Model of economic growth.  We consider the model of economic growth based upon three driving forces
namely, capital (including human capital), labor and technology. This is described as Neoclassical growth theory
in economics that layouts how steady economic growth rate can be accomplished with these three forces37–39.
The relationship between the labor and capital determines the output (i.e., the income), thus the model provides
a basis for economic growth in the canonical framework by defining a process that describes capital conversion
and accumulation in population over time40. This is a macro-oriented approach that dictate patterns of capital
accumulation as an inevitable outcome of normal market processes driven by savings and technological progress.
The patterns of accumulation converges over time leading to a steady economic growth across countries40.
We use a linear formulation of Constant-elasiticity-of-substitution (CES) production model, which is one of the
massively used function in economic analysis41. Thus, the rate of change of capital is given by:
dh
= {rhh + rl} − δ1h(t ),
dt (1)
where h(t) is the capital (income) at time t scaled to the labour supply. rh and rl determines the share in total out-
put from capital and labor respectively, so rh + rl = 1. δ1 is the rate of capital depreciation. Detailed discussion and
derivation of the linear form is given in the Supplementary Information (SI).

Model of strain dynamics: antibiotic resistance.  Acquired resistance may emerge from several bio-
logical mechanisms such as through mutation within the existing genome or through the plasmid transfer42.
For instance, resistance to the rifampicin drug in M.tuberculosis emerges through mutation whereas resistance
to antibiotic β-lactam in E. coli was by plasmid transfer43. However, the key issue in this epidemiological model
framework is to analyze the role of antibiotic consumption in the onset and propagation of resistance in the pop-
ulation. We assume that the drug-resistant strain is already present and circulating in the population. For sim-
plicity, we also assume that individual in population are either colonized by sensitive strains or by drug-resistant
strains. Colonization may result in the morbidity, and people undergo treatment also. We also consider that a sus-
ceptible individual may even consume antibiotics without having any prior knowledge of underlined infection,
but the drug has no effect if individual is colonized by resistant strain.
We consider a general SIS framework, where individuals become susceptible once recovered from an infec-
tion. Based on these assumptions, we compartmentalize the total population based on whether individuals are
using antibiotics or not, and further based on individual infection status – by the sensitive or the drug-resistant
strain. Figure 2 shows the schematic of the model and Table 1 provides description of the parameters used in the
model. Detailed description of the model is given in the Supplementary Information.
Transmission potential of the commensal bacteria is measured by basic reproduction number (o), which is
given by38:
β
o =
μ + γ1 (2)
In our model framework, resistant strains have a lower biological fitness as compared to the sensitive strains
in absence of drug. Irrational and extensive use of antibiotics exert a selective pressure by killing sensitive bacteria
and enable resistant bacteria to thrive and multiply44. The basic reproduction ′o for resistant strain is given by:

β′
′o =
μ + γ2 (3)
The dynamics of population infected by these two strains are determined by the parameters o and ′o.
We
shall assume that these are greater than unity (otherwise the commensal strains will never become established).
According to the ecological theory, individual strain is established in the population when the basic reproduction

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 3

www.nature.com/scientificreports/ www.nature.com/scientificreports

Parameters Description Values Reference

α̂ maximum antibiotic use 3, (1–5) calibrated
α
 minimum antibiotic use 0.01, (0.001–0.1) calibrated
rhc intrinsic growth rate of capital 0.9 56

rlc per capita amount spend on training or education of labor 0.8 56

ho education and awareness level 8, (7–15) calibrated

k proportion of individuals who are recovered due to drug use 0.1 calibrated
p proportion of individuals who use antibiotics even after recovery 0.1 calibrated
ρ proportion of infected (resistant) individuals take treatment 0.2 calibrated
δ1 rate of capital depreciation 0.1 56

δ2 average duration of antibiotic use 3/month 29

o reproduction number for sensitive strains 8, (7–9) 29

′o reproduction number for drug-resistant strain 10, (9–11) 29

γ1 recovery rate of sensitive strains 0.4/month 29

γ2 recovery rate of drug-resistant strains 0.3/month 29

cz per capita cost of treatment of severe infections 0.8 (0.1–2) calibrated

μ mortality rate 1
/year
72
Variables Description
s(t) Susceptible
S(t) Susceptible using antibiotics
y(t) Infective with sensitive commensals
z(t) Infective with resistant commensals
h(t) Capital or income

Table 1.  Description of variables and baseline parameter values (or ranges).

rate is greater than one, but strain with maximum basic reproduction number will always dominate over the other
strain. So here, the sensitive strain will dominate the population in absence of drug. Increased antibiotic con-
sumption however, can disrupt the balance in their reproductive ratios such that it reduces the fitness cost of
resistant strain and create a niche in the population of individuals treated with antibiotics.

Coupled disease-economic growth model.  In reality infectious disease and economic growth are
strongly coupled45–47. The correspondence between income and health is collective and circular causation: income
influences health and health influences income. Mechanism of infectious disease and economies are axiomatically
allied as poor health shrinks the economy by decreasing the capital and labor productivity, while the economic
wellness impacts the health like nutrition, sanitation, treatment, etc. Prevention and treatment of infections from
drug-resistant strains also thrust a strong financial burden worldwide, especially on low- and middle-income
countries by churning out huge sum of income or capital48.
We couple the two strains disease model and economic growth model by assuming that infections have det-
rimental impact on human capital accumulation and labor supply. It is based on recent literature studies in the
field of epidemiology and economics that human health have direct impact on cognitive, physical, and social
development which eventually have repercussion on economic development in future49–51. We consider that the
share parameters rh and rl in the economic growth equation (1) are functions of strains prevalences as follows:
rh(y , z ) = rhc(1 − y )(1 − z ) (4)

rl(y , z ) = rlc(1 − y )(1 − z ) (5)

where parameters rhc > 0 and, rlc > 0 determine the growth of capital in absence of any infection in the population.

Antibiotic misuse.  Antibiotic consumptions in developing and underdeveloped countries are strongly cor-
related with the economic growth of the populations. Several studies pointed out that poor economy fosters
self-medication practice in individuals, especially through over-the-counter (OTC) sales18,19,22. We develop a lin-
ear function of antibiotic use that depends on income on the basis of assumption that when income (capital) is
low, then self-medication is high and reaches a maximum level â. Similarly, when income is high then antibiotic
use reaches a minimum level a. The function is given by:
a(h) = mh + aˆ (6)
where slope of function m is determined by
a − aˆ
m= <0
ho (7)

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 4

www.nature.com/scientificreports/ www.nature.com/scientificreports

The parameter ho determines the effect of education and awareness of the community on antibiotic consump-
tion. A schematic of the function is given in Fig. S3 in SI. Figure S3 shows that higher the ho, lower the level of
awareness and hence higher the antibiotic consumption.
In reality, the association between the behavioral aspects of individual medical practice and socio-economic
growth of population is more complex and nonlinear, and there is rarely any empirical contribution in support of
this. Here, we intentionally consider a simpler approach, not to discount complexity, but to illustrate the complex
outcomes as emergent properties of parsimonious integrated models.

Integrated model and analysis.  We integrate the three components and construct the conceptual frame-
work that describes the feedback from one part to the other. The following coupled differential equations governs
the dynamics:
dS
= ahs + kahy − β ′Sz + pγ2z − δ2S − μS
dt
dy
= βsy − γ1y − kahy − μy
dt
dz
= β ′(s + S)z − γ2z − μz
dt
dh
= (rhh + rl ) − δ1h − ρczz
dt (8)

where ah = ( )
a − aˆ
ho
h + aˆ, and s + S + y + z = 1.
For simplicity, we assume that rh and rl are linear function of y and z as follows: rh(y , z ) ~ rhc(1 − y − z ),
rl(y , z ) ~ rlc(1 − y − z ). In the integrated model, we also assume that there is capital depreciation due to treat-
ment of infected individuals by resistant strain. The term ρczz denotes the depreciation, where ρ is probability of
maximum individual treated, and cz is per capita cost of treatment of severe infections. The variables and param-
eters descriptions are given in the Table 1 and in Supplementary Information.

Equilibrium solutions.  The dynamics of the system with either of strains (y = 0 or z = 0) is much simpler than that
of complete model given by (8). Here, we aim to discuss the effect of antibiotic consumption on the dynamics of both
sensitive and resistant strains in the population, and hence, mainly consider the existence of endemic equilibrium when
both pathogens exist in the system. Discussion of the endemic equilibrium provides us a framework to address interest-
ing issues such as invasiveness of resistance in the community, and also defines parametric regime related to antibiotic
use when resistant strains will come to dominate over sensitive strains, once present in the population.

Endemic equilibrium.  It is the condition where the commensals are not entirely eliminated from the community.
So, we have the susceptible individuals with and without antibiotic consumptions, and the infected individuals
either with sensitive or the resistant strain. The endemic equilibrium is given by
 1 ka
P ⁎(s ⁎, S ⁎, y ⁎ , z ⁎, h⁎):  + h , 1 − y ⁎ − s⁎
 o β
δ 2 + γ 2 (p − 1)
μ + γ2(1 − p) + − (ah + δ 2 + ′oμ + γ2(o′ − 1))s ⁎
⁎  ′o
−z , ,
kah + μ + γ2(1 − p) − (μ + γ2)o′ s ⁎
1 rlc − ρcz z ⁎ 0 ′ 
1 − y ⁎− , 
′o δ ′ − r c 
1 o h

Conditions for existence of endemic equilibrium. 

1. Sensitive strain: y ∗ > 0

kah
 ′0(ah +  ′0(μ + γ2 ) + δ 2 )[1 + ]
μ + γ1
                                if 0 > ′s(ah): =
μ  0′ + δ 2
The equilibrium density of sensitive strains y* is an decreasing function of antibiotic use ah. When ah
increases the threshold ′s(ah) also increases. This condition explains that the basic reproduction rate of
sensitive strain should be greater than the threshold value (′s(ah)) to persist in the population in presence
of resistant strain. Solving the inequality, the critical volume of antibiotic consumption to remove sensitive
strain is given by
2
1 1  A′ 1 1
ahc = B −  + 0 − B +  ,
4 C  C 2 C  (9)
kah A
where A, A′, B, C are given by A = μ′o + δ2, B = ′o(μ + γ2) + δ2, C = and A′ = . The detailed
μ + γ1 ′o
derivation of the threshold are given in the supplementary information.

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 5

www.nature.com/scientificreports/ www.nature.com/scientificreports

1
2. Resistant strain: z ⁎ > 0 if ′0 > ′r (ah): =
1 − y ⁎(ah)
For the resistance strain to invade, basic reproduction rate of resistance strain must be greater than the
threshold value ′r (ah), which is a decreasing function of ah.

(
3. Coexistence: 1co(ah): = 1 + kah
μ + γ1 ) 
< 0′ < co
0
2
(ah): =
δ2
1
(
1+
kah
μ + γ1 )
(ah + δ2 − γ2)

Condition (3) provides a parametric regime where both sensitive and resistant strains coexist in the system.
While 1co determines the density of susceptible population who are using antibiotics, the 2co gives a threshold
when the sensitive strain infects all individuals in the population. Both thresholds are increasing functions of ah,
and thus, the entire population will be infected by the resistant strain z(t) at higher ah.

Results
Conditions (1–3) discuss pathogen biology such as invasiveness and dominance of one strain on the other under
different parameters such as rate (ah) and duration (1/δ2) of antibiotic consumption, level of awareness and edu-
cation (h0). In absence of drug-resistant strain, the sensitive strain will be eradicated from the system if the anti-
biotic use reaches the threshold ahc given by equation (9) with ′0 = 1. An explicit expression is given in SI.
Prolong use of antibiotics will also accelerate this eradication process, as the threshold ′s is not only a function of
ahc , but also depends on δ2 i.e., the average duration of antibiotics consumption. However, in presence of
drug-resistant strain, the eradication of sensitive commensal is much easier, as the value of the threshold ahc
decreases in presence of resistant strain (condition 1). Higher the values of ′0, easier the eradication for sensitive
strain. Figure S4 in SI shows that the threshold ′s is increasing as the basic reproduction rate of resistant strain
′0 and antibiotic use ah is increasing. Thus, it will be easier to eradicate sensitive strain with high volume of anti-
biotic use in presence of resistant strain.
Nonetheless, invasion of drug-resistant strain in a community with prior colonized by sensitive strain requires
its basic reproduction rate is greater than ′r (condition 2), which decreases as the density of sensitive strain y*
decreases in the population. As ′r > 1, this signifies that colonization by drug-sensitive strain in the community
can inhibit invasion by drug-resistant strain provided that the level of antibiotic consumption remains below a
critical threshold ahc , and thus antibiotic consumption may escalate the invasion process of the resistant strain.
This shows an instance, where prior colonization of one form of commensal can protect the community from the
other one.
If drug-resistant strains have a transmission fitness advantage (which may be induced by antibiotic consump-
tion), then its population will grow and system will evolve towards coexistence. The relative fitness of both strains
determines the span of coexistence and it also depends on the niche created by antibiotic consumption rate ah in
population (condition 3). Increasing antibiotic consumption rate ah increases the range by extending 1co and 2co,
but eventually ′0 becomes greater than 0/1co , and all individual gets infected by the drug-resistant strain.
Figure S5 in SI shows the long-term dynamics of two strains under different values of relative fitness (′0/0) and
volume of antibiotic use (obtained using h0, see Fig. S3 for refertence). The resistant population grows and sensi-
tive population decreases as the relative fitness of resistant commensal increases, but the growth is much faster
when the rate of antibiotic consumption increases. At the higher relative fitness and higher values of antibiotic
use, only the resistant strain persist, wiping out the sensitive strain from the community.
In the numerical simulation, we mostly focus on the transient dynamics of the model system to see how the
integrated model (8) connecting three domains of human population, creates a reinforcing cycle to initiate the
development of resistance in population. Figure S6 in SI shows a sample time series explaining how the popula-
tion density of a highly prevalent commensal (′0 ~ 9) can successfully grow by intense antibiotic treatment
within a community. The eventual growth of resistance in the population has a negative impact on the economic
growth, leading to lower income and hence promotes more antibiotic consumption by self-medication, which in
turn, helps the resistance strain to grow and multiply towards stabilization. This phenomenon explains how three
components (Fig. 1) yield synergistic interactions forming a reinforcing cycle in developing drug-resistance in the
community.

Impact of economy and awareness on the development of resistance.  To investigate how

socio-economic growth factors can promote the drug-resistance in the population, we assume that the resistant
strain is already circulating in the population. In our model framework, we also differentiate populations based on
socio-economic growth characterized by two important parameters in our model: the initial value of the variable
h, i.e. h(0), which indicates economic status of the population in the beginning, and h0 which signifies education
and awareness-level of individuals in the population. For example, low h(0) represents a population with poor
economic status, but lower h0 represents relatively higher education and public health awareness in the population
(see Fig. S3 in SI). Thus, high h(0) together with lower h0 characterizes richer or developed countries, whereas
lower h(0) together with high h0 represents underdeveloped countries.
Figure 3(a) shows the time series of resistant strain z(t) for different values of h(0). As this shows, the popula-
tion with higher h(0) (representing relatively developed country), it takes 4 to 5 times longer duration to stabilize
the resistant commensal compare to population with lower h(0). Moreover, this has equal effect on the economy
and volume of antibiotic use by individuals in the population. This same scenario we have observed in many
countries today – countries with lower-middle (LMIC) and upper-middle (UMIC) income (Fig. S1) accumulate
more community-acquired resistance compared to developed countries8,52–54. To quantify the trajectory of devel-
opment of resistance along different income status, we plot the time of stabilization of resistant strain z(t), and
the area bounded by curves of h(t) and antibiotic use ah(t) for different values of h(0) (see Fig. S6 for reference).

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 6

www.nature.com/scientificreports/ www.nature.com/scientificreports

(a) (b)
1 80
Lower
income
0.8 70

time(month)
Stabilization
population
Resistant
0.6 60
Higher
0.4 income 50

0.2 40

0 30
0 20 40 60 80 0.05 0.5 1 1.5
(c)
Time (months) (d) Initial income (h(0))

250 200

150

consumption
200
reduction
Capital

Antibiotic
100

150
50

100 0
0.05 0.5 1 1.5 0.05 0.5 1 1.5
Initial income (h(0)) Initial income (h(0))

Figure 3.  Impact of economic status (h(0)) on the development of the drug-resistance. (a) Time series of
resistant strain for different values of h(0) - lower income drives early development of z(t). (b) Stabilization
time, (c) income reduction, and (d) increase in antibiotic consumption when h(0) varies 0.05–1.5. Low income
introduce more self medication, that drives early development of resistance, which in turn reduces more capital.
Drug-resistance develops relatively quickly in low-income countries and so higher income reduction. For detail
explanation, see the text.

We observe that there is steady increase in the time to stabilization (Fig. 3(b)), and steady decrease in the volume
of antibiotic consumption (Fig. 3(c)) as h(0) increases. However, the the amount capital reduction also decreases
with higher economy (Fig. 3(d)), but there is plateau showing less changes in capital reduction for low values of
h(0). Taken together, these results clearly explain how countries with lower economy can quickly develop a huge
burden of community-acquired drug-resistance by increasing the volume of antibiotic misuse, and damaging the
socio-economic growth.
We also experiment, in a similar way, the effect of awareness and education level on this dynamics, assuming
that education and awareness can significantly reduce the self-medication and over-the-counter (OTC) antibi-
otic consumption. Figure S7(a) shows that population with lower education level (higher h0) accumulates the
resistance much faster than population with higher education level (lower h0), though altogether the effect is less
compare to economy h(0) (compare Fig. 3(a)). The trajectory of stabilization time (Fig. S7(b)) decreases with
decrease in education level, and similarly the volume of antibiotic consumption (Fig. S7(c)) and capital reduction
(Fig. S7(d)) increases as education and awareness level decreases. Thus, better economic conditions and better
public health awareness may slow down the process of development of drug-resistant in population.
To characterize the relation between development of drug-resistance and antibiotic consumption, we also
plot the stabilization time of resistant strain with the volume and the duration of antibiotic use. As the Fig. 4
shows, stabilization of resistant strain varies significantly on the volume than the duration of consumption. It
is four-fold higher time to stabilize when volume of antibiotic consumption is very low. However, the duration
of consumption has greater impact when the volume of use is relatively lower. Hence, this indicates that public
health policy-makers requires to be more concerned on the per capita antibiotic use rather than the period of use.

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 7

www.nature.com/scientificreports/ www.nature.com/scientificreports

80

70

Stabilization time of
60

resistant strain
(months)
50

40

30

20
3
4
5
7
12
12 10.5
15 13.5
18 16.5
19.5
Volume of antibiotic

Figure 4.  Stabilization time of emergence of resistant strain under different volume and duration of antibiotic
consumption. The volume of antibiotic use corresponds to the area inside the triangle for each h0. When both
volume and duration are high, the resistance develops very quickly, but volume of consumption has larger
impact than the duration on the emergence process. For details, see the text.

Controlling the drug-resistance.  Our model demonstrates that community-acquired resistance may be
viewed as self-reinforcing process via antibiotic consumptions, which incurs huge economic burden to the soci-
ety. While the direct cost includes the prolonged hospitalization and treatment of patients, the indirect cost is
due to the productivity losses from excess morbidity and mortality attributable to resistant infections. To control
the situation, we require to slow down the process by perturbing the reinforcing cycle. One way to deal with this
is that government should provide monetary aid such as reducing the treatment cost or providing free health
service to the individuals infected by resistant strains. So, we experiment the effect of the monetary help on the
dynamics of drug-resistance by reducing the per capita cost of treatment cz in model equations (8). To replicate
more realistic scenario, we also consider that the rate ρ of maximum infected individuals are treated is a decreas-
ing function of cz by
ρ(cz ) = e−mc z, (10)
that is, more people look for treatment when cost of treatment is low. Figure 5(a) shows the impact of govern-
ment aid on the dynamics of resistance accumulation, capital reduction and antibiotic consumption over time.
In the simulation, we reduce the cost of treatment cz at the month five. The figure clearly shows that the popu-
lation colonized with resistant starts declining, capital starts increasing as the capital curve rises up and hence,
antibiotic consumption starts declining with the reduction in the cost of treatment. It is also observed that the
reduction in the cost of treatment lift the low-income countries back on the track of economic advancement.
Thus the amount of financial aid and timely initiative may help to decline the development of drug resistance
in population.
However, the actual reduction in the burden of resistance depends on when and how much aid is allo-
cated. To quantify this, we plot the area bounded by these two curves (before and after aid) in all three time
series for different values of amount of aid and time-to-aid. As observed in the Fig. 5(b), the situation is
much improved if the aid amount is higher and provided timely, especially, at the early stage of development
of drug-resistant in the community. Subsidized treatment cost or 100% aid will maximize the reduction of
resistance, but antibiotic mis(use) will drop to minimum at approximately 70% of aid. This analysis clearly
underscores that agencies and policymakers need to be more proactive to combat the situation of antibiotic
drug-resistance.

Implication of Model Prediction And Empirical Pattern

Our theoretical framework explores the development of community-acquired antibiotic resistance as a conse-
quence of antibiotic use, and discuss the implication of feedbacks in the dynamics from socioeconomic factors
such as income, education and awareness. The model puts together population ecology of infection disease, anti-
biotic consumption and economic growth forming a vicious cycle, and predicts that the population acquires high
burden of resistance in the long period of time that eventually leads the population to poverty. Empirical test-
ing of such feedback loops or parameter estimation of such reinforcing dynamics among different components
requires high-resolution data on transmission of resistance, antibiotic use, and impact of disease on the economic
growth at both population- and individua-level. To our knowledge, there is rarely such data available in public
repositories, and thus it becomes a serious challenge to test our theory. Instead, we allow our model to test for a
more approachable goal of determining the plausibility, that is, can our model predicts a long term trajectory to
reflect the empirical pattern observed in such population-level data?
We consider the data presented in Fig. S1 of the resistant prevalence of Klebsiella sp. and E. coli across dif-
ferent countries and the per capita gross national income (GNIP). As easily observed, the burden of resistance
is negatively correlated with the gross income across countries, and this highly nonlinear negative correlation

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 8

www.nature.com/scientificreports/ www.nature.com/scientificreports

Figure 5. (a) Impact of government aid to control the drug-resistance. While the dotted curve depicts the
dynamics of system before implementation of aid, the bold curve represents the effect after implementing the
aid by reducing the cost of treatment cz. The original cost of treatment is assumed cz = 12. The government
provides aid at the 5th month onward, then (i) curve of resistant population started declining, (ii) income curve
rises up (iii) antibiotic consumption curve goes down. (b) Plot of area bounded by the dotted and bold curves in
respective figures. Along with baseline values, other parameter values used for this simulation are
Ro = 7, δ1 = 0.11, δ2 = 2, ho = 7, aˆ = 2, a = 0.1, rhc = 0.9, rlc = 0.9, ρ = 0.07. For details, see the text.

suggests that there is a possible interaction between these two components through some feedback loop such as
one described above. To characterize empirical pattern in the data, we first perform a clustering analysis using
Gaussian Mixture Model (GMM), which is commonly used technique in statistical data analysis for grouping data
points or set of objects in such a way that data points in the same group are similar to one another in some sense
and distinct from point in other groups. Detailed discussion on algorithm and results from the analysis are given
in the supplementary information. The posterior probability obtained from the GMM indicates two components
in spatial distribution of the dataset, where contours define the measure of dispersion of data points from the cen-
troid of clusters under different confidence interval (Fig. 6). Results from the clustering analysis clearly illustrate
that the burden of resistance is higher in low-income countries (LIC), but it is lower in relatively high-income
countries (HIC).
We estimate model parameters such as the maximum and minimum antibiotic consumption level â and a,
duration δ2, education and awareness level ho, and the cost cz to treat per individual infected with the resistant
strain. As seen in both resistant pathogens Klebsiella sp. and E. coli, the parameter estimates for LIC determine
trajectory that approaches towards centroid of the clusters describing the higher burden of resistance, whereas the
model solution with HIC parameter values converges towards the cluster of lower disease burden (Fig. 6). In fact,
the basin of attraction for trajectories evolving towards centroid with higher disease burden can be seen as pov-
erty trap. The parameter estimates are given in Tables S3 and S4 in the supplementary information. However, this
model estimation supports the hypothesis that development of community-acquired drug-resistance can be seen
as a self-reinforcing system that increases the overuse and misuse of antibiotics and hinders socioeconomic
growth, eventually leading to poverty. While every country has its own economic constraints and epidemiological
parameters behind the selection and emergence of resistance, this analysis suggests that the data are consistent
with the theoretical framework considered here.

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 9

www.nature.com/scientificreports/ www.nature.com/scientificreports

(a) (b)
1 1

3GC resistant Klebsiella sp.

3CG resistant E. Coli.

0.1 0.1
contours of pdf
prevalence
HIC model
LIC model
centroid LIC
centroid HIC
0.01 0.01
3 4 5
10 10 10 10 3 10 4 10 5
Gross National Income per capita ($) Gross National Income per capita ($)

Figure 6.  Resistance and gross National income percapita (GNIP). (a) Clusters of Klebsiella sp. data and model
estimation. (b) Clusters of E. Coli data and model estimation. Refer the Fig. S1. The black cross mark and curve
represents the HIC and red cross mark and curve represents the LIC. For details, see the main text.

Discussion
An important paramount question in recent upsurge of interest in global health is why burden of antibiotic
drug-resistance follows very divergent trajectories in different parts of the world – it is converging to a controlled
situation in some parts, but ever-growing in the other parts like in India, China, and Sub-Saharan Africa! A paral-
lel predominant issue in global economy is also prevalent and increasing economic disparity between developed
and developing countries – that the rich continues to grow richer, and the poor gets poorer. While there are many
proposed explanations such as demographic55, political4, geographical9, and disease45,56 for this disparity from the
theory of economics and other disciplines, there are only a few that attempt to understand this from the perspec-
tive of worldwide scenario of antibiotic drug-resistance burden57.
We have hypothesized and devised a model of drug-resistance amalgamating population ecology of infectious
disease, socio-economic growth, and antibiotic (mis)use, that not only have the potential to demonstrate the
divergence in drug-resistance profile among different populations, but also explains the ever-growing disparity
in the global economy. The nonlinear interactions between income and health, and its consequence have been
pointed out in many literatures, but how feedbacks between economy and health can generate self-reinforcing
process by altering individuals behavior has been demonstrated in the present study. Empirical studies suggest
that individuals from LMICs choose self-medication to avoid high treatment cost, and thus pointing towards a
consistent increase in the gradient of antibiotic consumption. Our model also exhibits a similar pattern showing
a negative relationship of drug-resistance with gross income or capital of populations. Low education and lack of
awareness are other factors that promote self-medication. Currently, it has been observed that even the educated
population of developing countries are indulged into this defective practice such as in India19–21,58. We have also
shown using our model that low education and lack of awareness can also accelerate the emergence of resistant
strains and toll a significant economic cost on the population. Although the general relationship between the
burden of drug-resistance and economy of countries is interesting, it is more important to quantify the level of
economic growth beyond which the antibiotic consumption increases rapidly. Such results are useful to control
and policy interventions to manage antibiotic drug resistance. However, we need higher resolution data to quan-
tify such thresholds in this complex process.
Theory from physics states that a self-reinforcing process – which is made of chain of complex events – con-
tinue in the direction of their momentum until an external factor intervenes and breaks the cycle. Our simple
model also suggests that economic aid from government of countries or funding agencies can improve the present
situation of drug resistance by perturbing the self-reinforcing feedback loop, although a late response may not be
very effective. More the amount of aid and earlier the response, slower the development of resistance, and hence
better the economic growth in population. Thus, providing monetary aid in the form of reduced treatment cost,
inexpensive medicines and diagnostics are necessary for middle and low income countries, but that too before
the establishment of resistant strain in community. Development aids can yield everlasting economic and health
boon for the developing countries. Several international funding agencies such as GHIT, MOFA and MHLW in
Japan, and Bill & Melinda Gates foundation in US are working to combat drug-resistance in developing countries.
They are providing fund for development of new medicines, vaccines, and diagnostic tools to combat multi-drug
resistant disease like HIV/AIDS, tuberculosis, malaria59, although the impact and effectiveness of foreign aid in
developing countries depend on optimal allocation and coordination of the budget60.
While our theory provides a framework to explore such a complex system like antibiotic resistance, the model
has been simplified in several aspects. There are scopes for further improvement. We examine the model in the
absence of two biological mechanisms – mutation and plasmid transfer – which are important components for
the selection and emergence of multi-drug resistant pathogens in population. Also, the transmission mechanisms

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 10

www.nature.com/scientificreports/ www.nature.com/scientificreports

of different pathogens are different – for example, malaria (vector-borne), HIV/AIDS (exposure to body fluid)
and tuberculosis (Airborne). Our basic model may be improved and analyzed by integrating those specific char-
acteristics of the pathogens. The model can also analyze the effect of the poor drug quality, which is another crit-
ical issue in selection and emergence of drug-resistant strains61. Here we consider linear production production
function, which is a simpler approximation of actual constant-elasticity substitution (CES) function. Human
behavior and social interactions are other issues that influence the antibiotic consumption in community36. Our
study also motivates those who are readily curious to examine the interdependence of antibiotic resistance and
economic by incorporating other socio-economic parameters like nutrition, hygienic level, lifestyle, living condi-
tions, which serve as potential transmission factors of infection, along with migration, reproduction, and climate
change that may expand the practicality of the model.

Data Availability
Data is available as supplementary information.

References
1. Ventola, C. L. The antibiotic resistance crisis: part 1: causes and threats. Pharmacy and Therapeutics 40(4), 277 (2015).
2. Zaman, S. B. et al. A review on antibiotic resistance: alarm bells are ringing. Cureus 9(6) (2017).
3. Laxminarayan, R. et al. Antibiotic resistance: the need for global solutions. Lancet Infect Dis 13(12), 1057–1098 (2013).
4. Oneil, J. Review on antimicrobial resistance. Antimicrobial resistance: Tackling a crisis for the health and wealth of nations 2014
(2014).
5. Banga, J. M., Dube, S. & Munyanduki, H. Prevalence and drug resistance in bacteria of the urinary tract infections in bulawayo
province, zimbabwe. East Afr J Public Health 7(3) (2010).
6. Mason, P. R. et al. Antimicrobial resistance in gonococci isolated from patients and from commercial sex workers in harare,
zimbabwe. Int. J. Antimicrob. Agents 9(3), 175–179 (1998).
7. Williams, P. C. M., Isaacs, D. & Berkley, J. A. Antimicrobial resistance among children in sub-saharan africa. The Lancet Infect Dis
18(2), e33–e44 (2018).
8. Alvarez-Uria, G., Gandra, S. & Laxminarayan, R. Poverty and prevalence of antimicrobial resistance in invasive isolates. IJID 52,
59–61 (2016).
9. Klein, E. Y. et al. Global increase and geographic convergence in antibiotic consumption between 2000 and 2015. Proc. Nat. Acad.
Sci., page 201717295 (2018).
10. Woolhouse, M. & Farrar, J. Policy: an intergovernmental panel on antimicrobial resistance. Nature News 509(7502), 555 (2014).
11. Shrestha, P. et al. Enumerating the economic cost of antimicrobial resistance per antibiotic consumed to inform the evaluation of
interventions a_ecting their use. ARIC 7(1), 98 (2018).
12. Smith, R. & Coast, J. The true cost of antimicrobial resistance. Bmj 346, f1493 (2013).
13. Miller-Petrie, M., Pant, S. & Laxminarayan, R. Drug-resistant infections. In Major Infectious Diseases. 3rd edition. The International
Bank for Reconstruction and Development/The World Bank (2017).
14. Llorl, C. & Bjerrum, L. Antimicrobial resistance: risk associated with antibiotic overuse and initiatives to reduce the problem.
Therapeutic advances in drug safety 5(6), 229–241 (2014).
15. Holmes, A. H. et al. Piddock. Understanding the mechanisms and drivers of antimicrobial resistance. The Lancet 387(10014),
176–187 (2016).
16. World Health Organization et al. Global action plan on antimicrobial resistance. WHO report, accessed in January 2018, 2015.
17. Boeckel, P. V. et al. Global antibiotic consumption 2000 to 2010: an analysis of national pharmaceutical sales data. Lancet Infect Dis
14(8), 742–750 (2014).
18. Larissa Grigoryan, F. M. et al. Self-medication with antimicrobial drugs in europe. Emerging Infect. Dis. 12(3), 452 (2006).
19. Gebeyehu, E., Bantie, L. & Azage, M. Inappropriate use of antibiotics and its associated factors among urban and rural communities
of bahir dar city administration, northwest ethiopia. PLoS One 10(9), e0138179 (2015).
20. Planta, M. B. The role of poverty in antimicrobial resistance. JABFM 20(6), 533–539 (2007).
21. Barah, F. & Gonçalves, V. Antibiotic use and knowledge in the community in kalamoon, syrian arab republic: a cross-sectional study.
WHO report (2010).
22. Blenkinsopp, A. & Bradley, C. Over the counter drugs: Patients, society, and the increase in self medication. BMJ 312(7031),
629–632 (1996).
23. Islam, M. S. A review on the policy and practices of therapeutic drug uses in bangladesh. Calicut Med J 4(4), e2 (2006).
24. Yayehyirad, K. Self-care: a study of three communities in ethiopia. EJHD 2(2), 13–75 (1987).
25. Abula, T. & Worku, A. Self medication in three towns of north west ethiopia. Ethiop J (2001).
26. Shankar, P. R., Partha, P. & Shenoy, N. Self-medication and non-doctor prescription practices in pokhara valley, western nepal: a
questionnaire-based study. BMC family practice 3(1), 17 (2002).
27. Nguyen, Q. H. et al. Chuc. Unnecessary antibiotic use for mild acute respiratory infections during 28-day follow-up of 823 children
under five in rural vietnam. Trans. R. Soc. Trop. Med. Hyg. 105(11), 628–636 (2011).
28. Massad, E., Lundberg, S. & Yang, H. M. Modeling and simulating the evolution of resistance against antibiotics. International journal
of bio-medical computing 33(1), 65–81 (1993).
29. Austin, D. J., Kakehashi, M. & Anderson, R. M. The transmission dynamics of antibiotic– resistant bacteria: the relationship between
resistance in commensal organisms and antibiotic consumption. Proc R Soc Lond[Bio] 264(1388), 1629–1638 (1997).
30. Weinstein, R. A., Bonten, M. J. M., Austin, D. J. & Lipsitc, M. Understanding the spread of antibiotic resistant pathogens in hospitals:
mathematical models as tools for control. Clin. Infect. Dis. 33(10), 1739–1746 (2001).
31. DAgata, E. M. C., Webb, G. & Horn, M. A mathematical model quantifying the impact of antibiotic exposure and other interventions
on the endemic prevalence of vancomycin-resistant enterococci. IJID 192(11), 2004–2011 (2005).
32. Gagneux, S. et al. The competitive cost of antibiotic resistance in mycobacterium tuberculosis. Science 312(5782), 1944–1946 (2006).
33. Ankomah, P. & Levin, B. R. Two-drug antimicrobial chemotherapy: a mathematical model and experiments with mycobacterium
marinum. PLoS pathogens 8(1), e1002487 (2012).
34. Gumbo, T. et al. Selection of a moxioxacin dose that suppresses drug resistance in mycobacterium tuberculosis, by use of an in vitro
pharmacodynamic infection model and mathematical modeling. J. Infect. Dis. 190(9), 1642–1651 (2004).
35. Levin, B. R., Baquero, F. & Pål, J. J. A model-guided analysis and perspective on the evolution and epidemiology of antibiotic
resistance and its future. Current opinion in microbiology 19, 83–89 (2014).
36. Fu, F. & Chen, X. Social learning of prescribing behavior can promote population optimum of antibiotic use. Frontiers in Physics 6,
139 (2018).
37. Solow, R. M. A contribution to the theory of economic growth. QJE 70(1), 65–94 (1956).
38. Anderson, R. M. & May, R. M. Infectious diseases of humans: dynamics and control. Oxford university press (1992).

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 11

www.nature.com/scientificreports/ www.nature.com/scientificreports

39. Quah, D. T. Twin peaks: growth and convergence in models of distribution dynamics. The Economic Journal, pages 1045–1055
(1996).
40. Mankiw, N. G., Romer, D. & Weil, D. N. A contribution to the empirics of economic growth. QJE 107(2), 407–437 (1992).
41. Mc Faddenl, D. Constant elasticity of substitution production functions. The Review of Economic Studies 30(2), 73–83 (1963).
42. Tenover, F. C. Mechanisms of antimicrobial resistance in bacteria. AJICl 34(5), S3–S10 (2006).
43. Jacoby, G. A. Mechanisms of resistance to quinolones. Clin Infect Dis 41(Supplement_2), S120–S126 (2005).
44. Kolář, M., Urbánekl, K. & Látal, T. Antibiotic selective pressure and development of bacterial resistance. Int J Antimicrob Agents
17(5), 357–363 (2001).
45. Bonds, M. H., Keenan, D. C., Rohani, P. & Sachs, J. D. Poverty trap formed by the ecology of infectious diseases. Proc. R. Soc. Lond.
B 277(1685), 1185–1192 (2010).
46. Chakraborty, S. Endogenous lifetime and economic growth. JET 116(1), 119–137 (2004).
47. Bonds, M. H., Dobson, A. P. & Keenan, D. C. Disease ecology, biodiversity, and the latitudinal gradient in income. PLoS biology
10(12), e1001456 (2012).
48. Kawachi, I. & Kennedy, B. P. Socioeconomic determinants of health: Health and social cohesion: why care about income inequality?
BMJ 314(7086), 1037 (1997).
49. Nokes, C., Grantham-McGregor, S. M., Sawyer, A. W., Cooper, E. S. & Bundy, D. A. P. Parasitic helminth infection and cognitive
function in school children. Proc. R. Soc. Lond. B 247(1319), 77–81 (1992).
50. Miguel, E. & Kremerl, M. Worms: identifying impacts on education and health in the presence of treatment externalities.
Econometrica 72(1), 159–217 (2004).
51. Fernando, D. et al. A randomized, double-blind, placebo-controlled, clinical trial of the impact of malaria prevention on the
educational attainment of school children. ASTMH 74(3), 386–393 (2006).
52. Kesah, C. et al. Prevalence of methicillin-resistant staphylococcus aureus in eight african hospitals and malta. CMI 9(2), 153–156
(2003).
53. Shittu, A., Nübel, U., Udo, E., Lin, J. & Gaogakwe, S. Characterization of meticillin-resistant staphylococcus aureus isolates from
hospitals in kwazulu-natal province, republic of south africa. JMM 58(9), 1219–1226 (2009).
54. Onipede, A. O. et al. Invasive bacteria isolates from children with severe infections in a nigerian hospital. JIDC 3(06), 429–436
(2009).
55. Burtless, G. Demographic transformation and economic inequality. The Oxford Handbook of Economic Inequality (2009).
56. Ngonghala, C. N. et al. Poverty, disease, and the ecology of complex systems. PLoS biology 12(4), e1001827 (2014).
57. Cockburn, R., Newton, P. N., Agyarko, E. K., Akunyili, D. & White, N. J. Correction: the global threat of counterfeit drugs: why
industry and governments must communicate the dangers. PLoS medicine 4(9), e289 (2007).
58. Rather, I. A., Kim, B. C., Bajpai, V. V. & Park, Y.-H. Self-medication and antibiotic resistance: Crisis, current challenges, and
prevention. Saudi J Biol Sci 24(4), 808–812 (2017).
59. Sachi Gerbin, C. Enhancing us-japan cooperation to combat antimicrobial resistance. Biosecur Bioterror 12(6), 337–345 (2014).
60. Rajlakshmi D. & Charles B. The foreign aid effectiveness debate: Evidence from malawi 6 (accessed in April 2019), http://docs.
aiddata.org/ad4/pdfs, (2015).
61. Pisani, E. Antimicrobial resistance: What does medicine quality have to do with it. Review on Antimicrobial Resistance (2015).

Acknowledgements
Authors thank all the reviewers for their comments and suggestions that greatly improved the presentation of
the manuscript. Authors also thank DST, Science and Engineering Research Board (SERB), India for financial
support (ECR/2016/000591) to conduct this research.

Author Contributions
Conceived and designed the study: S.B. Preparation of study materials: B.M. Performed analysis: B.M. and S.B.
Wrote the manuscript: S.B. and B.M. All authors approved the final draft.

Additional Information
Supplementary information accompanies this paper at https://doi.org/10.1038/s41598-019-46078-y.
Competing Interests: The authors declare no competing interests.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Cre-
ative Commons license, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons license and your intended use is not per-
mitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

© The Author(s) 2019

Scientific Reports | (2019) 9:9788 | https://doi.org/10.1038/s41598-019-46078-y 12