DTIC ADA593313: Risk-based Classification System of Nanomaterials

0.27 MB·English

by  Defense Technical Information Center

JNanopartRes
DOI10.1007/s11051-008-9546-1
RESEARCH PAPER
Risk-based classification system of nanomaterials
Tommi Tervonen Æ Igor Linkov Æ
Jose´ Rui Figueira Æ Jeffery Steevens Æ
Mark Chappell Æ Myriam Merad
Received:13April2008/Accepted:18October2008
(cid:1)USGovernment2008
Abstract Various stakeholders are increasingly nanomaterial production. To guide scientists and
interested in the potential toxicity and other risks engineers in nanomaterial research and application
associated with nanomaterials throughout the differ- as well as to promote the safe handling and use of
entstagesofaproduct’slifecycle(e.g.,development, thesematerials,weproposeadecisionsupportsystem
production, use, disposal). Risk assessment methods for classifying nanomaterials into different risk cate-
and tools developed and applied to chemical and gories. The classification system is based on a set of
biological materialsmay not be readily adaptable for performance metrics that measure both the toxicity
nanomaterials because of the current uncertainty in and physico-chemical characteristics of the original
identifyingtherelevantphysico-chemicalandbiolog- materials, as well as the expected environmental
icalpropertiesthatadequatelydescribethematerials. impacts through the product life cycle. Stochastic
Such uncertainty is further driven by the substantial multicriteria acceptability analysis (SMAA-TRI), a
variationsinthepropertiesoftheoriginalmaterialdue formal decision analysis method, was used as the
to variable manufacturing processes employed in foundation for this task. This method allowed us to
cluster various nanomaterials in different ecological
T.Tervonen J.Steevens(cid:1)M.Chappell
FacultyofEconomicsandBusiness,Universityof USArmyResearchandDevelopmentCenter,
Groningen,P.O.Box800,9700AVGroningen, CEERD-EP-R,Vicksburg,MS39180,USA
TheNetherlands
e-mail:[email protected] J.Steevens
e-mail:[email protected]
I.Linkov(&)
USArmyResearchandDevelopmentCenter, M.Chappell
83WinchesterStreet,Suite1,Brookline,MA02446, e-mail:[email protected]
USA
e-mail:[email protected] M.Merad
SocietalManagementofRisksUnit/AccidentalRisks
J.R.Figueira Division,INERISBP2,60550Verneuil-en-Halatte,
CEG-IST,CentreforManagementStudies,Instituto France
SuperiorTe´cnico,TechnicalUniversityofLisbon, e-mail:[email protected]
2780-990PortoSalvo,Portugal
e-mail:fi[email protected]
J.R.Figueira
LAMSADE,Universite´ Paris,Paris,France
123
Report Documentation Page Form Approved
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.
1. REPORT DATE 3. DATES COVERED
OCT 2008 2. REPORT TYPE 00-00-2008 to 00-00-2008
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Risk-based Classification System Of Nanomaterials
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
US Army Research and Development Center,83 Winchester Street, Suite REPORT NUMBER
1,Brookline,MA,02446
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
Journal of Nanoparticle Research
14. ABSTRACT
Various stakeholders are increasingly interested in the potential toxicity and other risks associated with
nanomaterials throughout the different stages of a product?s life cycle (e.g., development, production, use,
disposal). Risk assessment methods and tools developed and applied to chemical and biological materials
may not be readily adaptable for nanomaterials because of the current uncertainty in identifying the
relevant physico-chemical and biological properties that adequately describe the materials. Such
uncertainty is further driven by the substantial variations in the properties of the original material due to
variable manufacturing processes employed in nanomaterial production. To guide scientists and engineers
in nanomaterial research and application as well as to promote the safe handling and use of these
materials, we propose a decision support system for classifying nanomaterials into different risk categories.
The classification system is based on a set of performance metrics that measure both the toxicity and
physico-chemical characteristics of the original materials, as well as the expected environmental impacts
through the product life cycle. Stochastic multicriteria acceptability analysis (SMAA-TRI), a formal
decision analysis method, was used as the foundation for this task. This method allowed us to cluster
various nanomaterials in different ecological risk categories based on our current knowledge of
nanomaterial physico-chemical characteristics, variation in produced material, and best professional
judgments. SMAA-TRI uses Monte Carlo simulations to explore all feasible values for weights, criteria
measurements, and other model parameters to assess the robustness of nanomaterial grouping for risk
management purposes.
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON
a. REPORT b. ABSTRACT c. THIS PAGE Same as 11
unclassified unclassified unclassified Report (SAR)
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
JNanopartRes
risk categories based on our current knowledge of for nanomaterials that takes into account several
nanomaterial physico-chemical characteristics, varia- nanomaterial parameters commonly associated with
tion in produced material, and best professional ecotoxicityandenvironmentalrisk.Theseparameters
judgments.SMAA-TRIusesMonteCarlosimulations vary from nanomaterial physico-chemical character-
to explore all feasible values for weights, criteria isticstoexpectedenvironmentalconcentrationstofate
measurements, and other model parameters to assess and transport mechanisms. We focus primarily on
the robustness of nanomaterial grouping for risk ecological risks, although the same methodology
management purposes. could be applied to human health risk assessment.
Thisworkdoesnotattempttodrawexactconclusions
Keywords Nanotechnology (cid:1) Risk assessment (cid:1) about the environmental risks associated with differ-
Toxicology (cid:1) Decision analysis (cid:1) Governance ent nanomaterials, but rather to provide reasonable
recommendations about which nanomaterials may
need more precise measurements and testing to be
safely deployed inconsumer products.
Introduction
Nanotechnologyisarapidlygrowingfieldofresearch MCDA approaches to classification
that is already having a great impact on consumer
products.Thefieldofnanotechnologycanbedefined Clustering nanomaterials into ordered risk categories
astheproductionanduseofmaterialsatthenanoscale, can be treated as a sorting problem in the context of
normallycharacterizedassmallerthan100 nminone multi-criteria decision analysis (MCDA). MCDA
dimension (Oberdo¨rster et al. 2007). Nanomaterials refers to a group of methods used to impart structure
areformedthroughbothnatural(e.g.,combustionby- to the decision-making process. Generally, the
products)andsyntheticprocesses.Forthepurposesof MCDA process consists of four steps: (1) structuring
this article, we focus our discussion solely on the problem by identifying stakeholders and criteria
engineered nanomaterials, which are currently used (nanomaterial properties in this case) relevant to the
inmorethan600differentconsumerproducts(Wood- decision at hand, (2) eliciting the parameters of the
rowWilsonInstitute,Onlinedatabase,2008).Inspite model (weights, thresholds, etc.) and assigning mea-
of their potential commercial benefits, some nanom- surementsforeachalternative(e.g.,nanomaterialrisk
aterials have been identified as toxic in in vivo and group), (3) executing the model through computer
in vitrotests.Clearly,ourknowledgeofthepotential software,and(4)interpretingresultsofthemodeland
toxicity of these materials is far from comprehensive possiblyre-iteratingtheprocessfromstep1orstep2
(Oberdo¨rsteretal.2005;ThomasandSayre2005).The by re-evaluating the model. The goal of this MCDA
potential environmental fate and toxicity (as well as processisnottoselectasinglebestalternative,butto
potentialforexposureandrisk)ofnanomaterialsmay rank or group alternatives through a structured
be strongly impacted by the material’s physico- process. A detailed analysis of the theoretical foun-
chemical characteristics. For example, potentially dations for different MCDA methods and their
toxic nanoparticles (NPs) that tightly bind to soil comparative strengths and weaknesses is presented
surfaces may exhibit limited movement through the in Belton and Stewart (2002). A review of MCDA
environment. In this case, such materials may be applications to environmental management can be
deemed relativelysafe for certain specificuses. Such found in Linkov et al. (2006); risk-based decision
informationisimportantasalackofunderstandingof frameworkforselectingnanomaterialforspecificuse
nanomaterial toxicity and risk may delay full-scale is discussed in Linkov et al. (2007).
industrialapplicationofnanotechnologies. The SMAA-TRI sorting method (Tervonen et al.
Nanomaterial research and regulations could be 2009) is well suited for the proposed classification
guided by a systematic characterization of factors systemgiventheuncertaintyofavailableinformation
leading to toxicity and risks in the absence of regarding the physico-chemical characteristics of
definitivedata(LinkovandSatterstrom2008).Inthis nanomaterials(seeFigueiraetal.2005a,forareview
article, we propose a risk-based classification system of other MCDA sorting methods). Many of the
123
JNanopartRes
characteristicsattributedtonanomaterialsarelimited classifications because the weights represent ‘‘votes’’
to a solely qualitative assessment. We used SMAA- for each criterion which are not affected by criteria
TRI, an outranking model based on ELECTRE TRI scales. The lambda cutting level represents the
(see e.g., Figueira et al. 2005b) for the assignment minimum weighted sum of criteria that have to be
procedure. If an alternative outranked another, then in concordance with the outranking relation for it to
the alternative was considered at least as good as or hold:thelambdacuttinglevelisusedtotransformthe
betterthan another alternative.We preferred SMAA- ‘‘fuzzy’’ outranking relation into an exact one
TRI, since it extends the capabilities of ELECTRE (whether an alternative outranks a profile or not).
TRI by allowing the use of imprecise parameter Forexample,alambdacuttinglevelof0.6meansthat
values. ELECTRE TRI assigns the alternatives (dif- 60% of the weighted criteria have to be ‘‘at least as
ferent nanomaterials in this study) to ordered good’’ for the outranking relation to hold.
categories (risk classes). Three types of thresholds Alternatives were compared by accounting for the
are used to construct the outranking relationships by three thresholds. An alternative and profile with
definingpreferenceswithrespecttoasinglecriterion. scores of 0.4 and 0.6 (for the same criterion),
Theindifference thresholddefines thedifferenceina respectively, and anindifference threshold ofat least
criterion that is deemed insignificant. The preference 0.2 demonstrates that this criterion fully supports the
thresholdisthesmallestdifferencethatwouldchange conclusion that the alternative outranks the profile.
the expert preference. Between these two lay a zone Sometimes the support is not binary but is further
of ‘‘hesitation’’ or indifference. The veto threshold is affected by linear interpolation inthe hesitation zone
the smallest difference that completely nullifies of both veto and preference thresholds (see e.g.,
(raises a ‘‘veto’’ against) the outranking relation. Tervonen2007).Inthiscasethesupportcanhavereal
The assignment procedure involves comparing the values between 0 (no support) and 1 (full support).
propertiesassociatedwithaspecificnanomaterial(g , All the parameters of ELECTRE TRI can be
1
g , …, g ) against a profile that includes ranges of imprecise and represented by arbitrary joint distribu-
2 m
criteria metric values corresponding to several risk tions in SMAA-TRI. This feature allows us to make
classes. Comparisons are performed with respect to conclusions about risks related to different nanoma-
each criterion, taking into account the specified terials even though the information about their
thresholds. The final classification decision is based characteristics is limited. Monte Carlo simulations
on the profile criteria weights and specified cutoff were used in SMAA-TRI to compute acceptability
level (lambda). For example, Class 4 represents the indices for alternative categorizations (i.e., for
highest risk while Class 1 is the lowest risk (Fig. 1). assigning nanomaterials in different risk classes).
The assigned criteria weights represent the sub- SMAA-TRI allows automatic sensitivity analysis.
jective importance of the criteria. For this reason, SMAA-TRI output comes as a set of category
ELECTRE TRI was particularly attractive for these acceptability indices which describes the share of
feasible parameter values that assign alternatives to
each category. Thecategory acceptabilityindices are
g
m measures indicating the stability of the parameters,
g
m-1
i.e., if the parameters are too uncertain to make
informed decisions. A high index ([95%) signals a
reasonablysafeassignmentofthealternativeintothe
Class 4 Class 3 Class 2 Class 1 corresponding category. With lower indices, the risk
attitude of the decision maker defines the final
assignment. For example, if an alternative has an
80% acceptability for the lowest risk category and a
g
g3 20% acceptability for the second lowest risk cate-
2
g1 gory, a risk-averse decision maker could assign the
alternative to the higher risk category.
Fig.1 Examplemeasurementsofprofilesforeachcriteriong
j SMAA-TRI conducts the numerical simulation by
(adapted from Merad et al. 2004). Profiles are marked with
horizontallines comparing the effects of changing parameter values
123
JNanopartRes
and criteria evaluations on the modeling outcomes. gravity, as described by Stokes law relationships.
Parameter imprecision can be quantified by Monte Particles may settle but remain non-flocculated,
Carlo simulations using different probability distri- settling at interparticle distances with the lowest
butions(uniform,normal,log-normal,etc.).Gaussian free energies. In the absence of surfactive agents,
or uniform distributions are typically used (for more particle flocculation is fairly predictable using
informationaboutSMAAmethods,seeTervonenand particle charge. Charged functional groups give
Figueira 2008). way to the development of a surface electrostatic
If some model parameters need to have their potential which extends out a few nanometers at
sensitivityassessed,theycanbeconsideredimprecise the solid–liquid interface, forming a diffuse
and defined as probability distributions. doublelayerorDDL(Bowdenetal.1977;Uehara
and Gillman 1981). Classical DLVO theory
predicts that repulsive forces between particles
(arising from overlapping DDLs) increase with
Criteria
increasing ion concentrations (or increasing ionic
strength, I)becauseofrising osmotic pressuresat
Recent articles, as well as the frameworks reviewed
the solid–solution interface force the DDL to
in this study, generally propose several different
swell (Evangelou 1998, and references therein).
characteristics in the risk assessment of nanomate-
Yet, classical Debeye–Huckel theory predicts a
rials. These characteristics are generally based on
competing case where increasing ion concentra-
extrinsic particle characteristics (size, agglomera-
tion decreases DDL thickness, throwing a system
tion, surface reactivity, number of critical function
intoflocculation.Thus,atafundamentallevel,the
groups, dissociation abilities) (Biswas and Wu 2005;
process of agglomeration represents the balance
Borm and Mu¨ller-Schulte 2006; Borm et al. 2006;
of these two competing charge interactions.
Gwinn and Vallyathan 2006; Kreyling et al. 2006;
• Reactivity/charge: A NP may become charged
Medina et al. 2007; Nel et al. 2006; Oberdo¨rster
either by design (such as through functionaliza-
et al. 2005; Thomas and Sayre 2005). These various
tion) or by spontaneous degradative reactions.
parameters are critical because they define the fate
NPs may be functionalized with various types of
and relevant intact exposure pathways as well as
groups, such as COOH, NH , and SH through
internal dose required to assess risk (Tsuji et al. 2 2
standard organic synthesis methods. Such func-
2006). Summary descriptions of five basic extrinsic
tionalizations may be useful for manufacturing
nanomaterial properties (agglomeration and aggre-
processes. For example, single-walled carbon
gation, reactivity, critical functional groups, particle
nanotubes (SWNTs) are typically carboxylated
size, and contaminant dissociation) are presented
at their ends as part of the isolation/purification
below:
processes (Anita Lewin, RTI International, per-
• Agglomeration, weakly bound particles, and sonal communication). The type of charge
aggregation, strongly bound or fused particles occurringonfunctionalizedNPsiscalledvariable
(ISO 2008), are important risk criteria because charge, which means that the magnitude of the
theyprovide adescriptionofthephysicalstateof surfaceelectrostaticpotentialvarieswithsolution
NPs in the aquatic system (Kennedy et al. 2008; pH(UeharaandGillman1981).Variablycharged
Wang et al. 2008). In aqueous solutions, groups characteristically exhibit a surface pK .
a
NP agglomeration generally occurs by two Thus, variably charged surface groups may be
mechanisms: colloid settling and flocculation. speciated (e.g., protonated versus deprotonated)
Flocculation occurs when Brownian-driven colli- by the classical Henderson–Hasselbauch equa-
sionsbindunassociatedparticlestogetherthrough tion. Furthermore, the magnitude of the surface
Van der Walls forces by dehydrating the inter- electricalpotentialmaybesuppressedbyincreas-
acting surfaces. Consequently, the particle ingI,asdescribedpreviously.Thus,thereactivity
separates out of the solution containing the mass of variably charged functional groups varies with
of the previously unassociated particles. Settling, thedifferenceinsolutionpHfromthesurfacepK
a
on the other hand, occurs due to the pull of and the magnitude of I.
123
JNanopartRes
• Critical functional groups: Related to the reactiv- accompanying bulk have been shown to possess
ity/charge, critical functional groups make up an enhanced reactivities, such as high-affinity
important criterion given the fact that nanomate- adsorption of metals or unique structures of
rial functionality and bioavailability is directly assembly during agglomeration (Auffan et al.
relatedtochemicalspecies.Basingriskcriteriaon 2008; Erbs et al. 2008). Particle size is particu-
elemental speciation is superior to elemental larly important in terms of distinguishing the
compositionalonebecauseitidentifiestheunique unique size-dependent chemistry of NPs from
set of reactions available to each species. For classical colloid chemistry.
example, suspended zero-valent Fe NPs have
Factorsthatmayinfluencethepotentialhazardsof
been shown to catalyze reductive degradations of
engineered nanomaterials include bioavailability,
aqueous organic contaminants (Joo et al. 2004).
bioaccumulation and translocation potential, and
The same degradative ability has been shown for
potential for toxicity. These factors have been
structural Fe2? (higher oxidation state than zero-
described in empirical studies and are dependent on
valent Fe but different speciation in terms of its
thecharacteristicsoftheparticlesasdescribedabove.
complexation environment) domains at clay-edge
Itisdifficulttopredictthebehaviorofnanomaterials;
and -interlayer nanosites in soil (Hofstetter et al.
however, future computational approaches are
1999, 2003). The Cd2? cation in quantum dots
expectedtoprovideadditionaltoolstoestimatethese
exhibits no toxicity to organisms as long as it
properties from physical and chemical parameters.
remains complexed with Se (Derfus et al. 2004).
Speciation also determines solubility or potential • Bioavailability:Bioavailabilitydescribesthelike-
dissociation of nanomaterials. lihood of a material to be absorbed across cell
• Contaminantdissociation:Thiscriteriondescribes membranes from the various exposure routes
risk associated with residual impurities contained (e.g., dermal, inhalation, oral exposures) into
within the NP. For example, Fe oxide NPs may system circulation inan organism (Medinsky and
contain S impurities depending on whether FeCl Valentine2001).Thisprocessiscontrolledbythe
3
or Fe (SO ) was used in manufacturing. Carbon characteristics described above. For example,
2 4 3
nanotubes may contain Ni, Y, or Rb metal cation particle charge may influence agglomeration and
impurities (Bortoleto et al. 2007; Chen et al. hence limit the ability of the particle to cross
2004), which may either be entrained within or gastrointestinal membranes after oral ingestion.
adsorbed onto the surface of the tubes. However, However, several pathways enable NPs to cross
little is actually known about the extent to which cellmembranes,includingpinocytosis,endocyto-
metallic and organic contaminants remain with sis,anddiffusion(assummarizedbyUnfriedetal.
the manufactured product. Thus, the assignment 2007). The mechanism by which particles are
of this risk criterion could change with better absorbed is highly dependent on particle compo-
information. sition, surface modification, size, shape, and
• Size: Particle size is a criterion related to the agglomeration.
agglomeration and reactivity criteria. Obviously, • Bioaccumulation potential: Bioaccumulation is
smaller particles agglomerate at slower rates. the net accumulation of particles absorbed from
However, agglomeration is also related to the all sources (soil, water, air, and food) and
particle size distribution or polydispersivity. For exposure routes listed above into an organism.
example, greater monodispersivity of particles Accumulationmustconsiderthetemporalaspects
sizes appears to promote more stable dispersions of exposure and include kinetic factors such as
(Chappellet al. 2008).Also, NPreactivityisalso exposure concentration, duration of exposure,
impactedbythemagnitudeofNPsurfacerelative clearance, biotransformation, and degradation.
to the bulk of the solid. While the surface is the Moststudiestodatehavefocusedonthepotential
reactive portion of solids, the bulk component for uptake and translocation in specific tissues
may suppress the surface reactivity through (Ryman-Rasmussen et al. 2006; Gopee et al.
internal reorganizations, etc. NPs are essentially 2007; Kashiwada 2006) and have not addressed
surfaces with limited bulk. Surfaces with low the toxicokinetics of NPs.
123
JNanopartRes
• Toxic potential: The toxicity of engineered number was preferred to a larger one, but the
nanomaterials and particles in mammalian and intervals did not carry any information (e.g., 1 is as
other animal systemshas been assessed primarily much preferred to 2 as 1 is to 3). If there were
through cytotoxicity screening assays, although multiple possible classes for an alternative, the
some in vivo studies have been completed. It is measurement was modeled with a discrete uniform
proposed that toxicological effects of nanomate- distribution,meaningthatthedensityfunctionforthe
rials occur throughoxidativestress,inflammation distribution was such that the integers corresponding
from physical irritation, dissolution of free metal to these classes were equiprobable. Veto thresholds
from metal NPs, and impurities in nanomaterials werenotusedinthisphaseoftheframeworkbutwill
(e.g., catalysts) (Oberdo¨rster et al. 2007). The be added later when more information about the
characteristics of NPs that influence toxicity criteria becomes available. Size is a criterion that
include size, surface area, morphology, and shouldhavesomevetoassociatedwithitsothatvery
dissolution. To date, screening studies using smallmaterialscannotbeassignedtothesafer(lower
in vitro approaches have observed toxicity from risk) categories.
metal NPs at lower concentrations (Braydich- Even though nanomaterial size is believed to be a
Stolleetal.2007)thantoxicityfromcarbon-based factor influencing toxicity, there is little specific
NPs (Murr et al. 2005; Grabinski et al. 2007). information available characterizing toxic effects
relative to the 1–100 nm size range (Powers et al.
2007). More research is needed to define the thresh-
oldsinamoreexactmanner.Ifa‘‘smaller’’-sizedNP
represents higher risk, it follows that a larger size is
Proposed classification framework ‘‘more preferable’’ because of its inherently lower
risk. Due to these knowledge gaps, imprecise thresh-
The purpose of the proposed classification system is olds were used for nanomaterial size with
to preliminarily group nanomaterials in risk classes indifference threshold of 10 ± 5% and preference
for screening level risk assessments. Such groupings threshold of 25 ± 5%.
should aid in prioritizing materials for further study. Bioavailability, bioaccumulation, and toxic poten-
In this article, we considered five risk categories: tialwereallmeasuredusingacardinalbutsubjective
extreme, high, medium, low, and very low risk. In scale as described above. Because of the subjectivity
order to assign particular nanomaterials to these of this scale, we applied imprecise thresholds.
categories, we need to define criteria scales, thresh- Indifference thresholds were set to vary uniformly
olds, and measurements (Table 1). from 0 to 10, and preference thresholds from 10 to
The quantitative criterion, particle size, was eval- 20.
uated as the mean size of the material in units of The SMAA-TRI model separated the risk catego-
nanometers as obtained from literature review and ries using profiles formed from measurements of the
expert estimates. Bioavailability, bioaccumulation, same criteria as the alternatives. In our framework,
andtoxicpotentialweremeasuredthroughsubjective the profile measurements were all exact (Table 2).
probabilities that the nanomaterial has significant Our model applied imprecise preference infor-
potentialinthecriterion.These,aswellastherestof mation in the form of weight bounds. For more
the criteria (agglomeration, reactivity/charge, critical information on how these were implemented, see
function groups), were measured based on expert TervonenandLahdelma(2007).Wejudgedthetoxic
judgments. The qualitative criteria were measured in potential to be the most important criterion, and thus
terms of ordinal classes: 1 was the most favorable it was assigned weight bounds of 0.3–0.5. Bioavail-
(least risk) value class, while 5 the least favorable ability and bioaccumulation potentials were deemed
(highest risk) (Table 1). the least important criteria, and as a result, we were
Forthequalitativecriteria,weencodedtheclasses undecidedontheirrelativeimportance.Bothofthese
withintegers.Theindifferencethresholdsweresetto criteria were given weight bounds ranging from 0.02
0 and the preference thresholds to 1. This choice of to 0.08. The rest of the criteria were assigned weight
thresholds represented an ordinal scale: a smaller bounds of 0.05–0.15.
123
JNanopartRes
Table1 Criteriameasurements
Agglomeration Reactivity/ Critical Contaminant Bioavailability Bioaccumulation Toxic Size
charge function dissociation potential(±10) potential(±10) potential (±10%)
groups (±10)
C60 4 2,3 3 2 25 50 10 100
MWCNT 4 2,3 4 3 25 50 25 50
CdSe 4 4,5 1 4 50 75 75 20
AgNP 3 4,5 1 4 50 75 75 50
AlNP 5 1,2 1 1 25 75 10 50
The first four criteria are measured as ordinal classes. Measurements of reactivity/charge have associated uncertainty in that the
materialscanbelongtoeitheroftheindicatedclasses.Thefollowingthreecriteriahavelinearimprecisionof10inbothdirections
fromtheindicatedmeanvalue.Sizehasuncertaintyof10%oftheshownmeanvalue
Table2 Profilemeasurements
Profile Agglomeration Reactivity/ Critical Contaminant Bioavailability Bioaccumulation Toxic Size
charge functiongroups dissociation potential potential potential
Extreme– 4 4 4 4 100 100 100 5
high
High– 3 3 3 3 80 80 80 50
medium
Medium– 2 2 2 2 70 70 70 100
low
Low–very 1 1 1 1 60 60 60 200
low
Eachrowcorrespondstoaprofiledifferentiatingthecategoriespresentedinthefirstcolumn
We used imprecise values for the lambda cutting each criterion described in ‘‘Criteria’’ section. Met-
level within the range of 0.65–0.85. Lambda defines rics for the five materials used in our case study
theminimumsumofweightsforthecriteriathatmust (Table 2) as well other model parameters were input
be in concordance with the outranking relation to into the SMAA-TRI software. Even though criteria
hold. The classification was performed according to metricsusedinthearticlewereassessedusingexpert
the pessimistic assignment rule, which in risk judgment, and its objectivity can be questioned, the
assessment applications represents a more conserva- outranking algorithms used in SMAA-TRI together
tive approach. with the choice of absolute thresholds implemented
in this study allows us to obtain robust results
(Tervonen 2007).
Example Category acceptability indices obtained from the
simulation are presented in Fig. 2. These indices
We demonstrated application of the framework by showthatthedatawastooimprecisetomakedefinite
classifying five nanomaterials: nC (a fullerene), decisions about the risks related to the different
60
multi-walled carbon nanotube (MWCNT), CdSe nanomaterials. However, there was sufficient data to
(quantum dot), silver nanoparticles (Ag NPs), and make preliminary classifications. For example, CdSe
aluminum nanoparticles (Al NPs). Typical size exhibitedaveryhighindexinthehighriskclass.On
ranges for these materials were estimated based on the other hand, Al NP may be considered relatively
in situ measurements from the available literature. safe; its category acceptability indices for low and
Other properties were assessed using authors’ expert verylowriskwere34and34,respectively.Summing
judgments, taking into accountthe characteristics for these indices gave the material an estimated 68%
123
JNanopartRes
highest risk class potentially represent areas of
importantfuturetoxicologicalstudies,whilematerials
exhibitinglowriskmayberecommendedastargetsof
research aimed at commercial use. The proposed
framework takes into account measurements and
expert estimates for multiple criteria that are known
toimpact the toxicityof the material.
TheuseofanSMAA-TRIapproachallowsforthe
explicit incorporation of uncertainty parameters in
the model. An appealing characteristic of the outran-
kingmodelappliedinSMAA-TRIisthatitallowsthe
Fig.2 Categoryacceptabilityindicesoftheexample.Ahigh
veto effect to be modeled, meaning that a nanoma-
indexmeansthatthematerialisassignedtothecorresponding
category with a high confidence, as measured by a larger terial’s poor performance in one criterion cannot be
percentageshareofpossibleparametervaluescorrespondingto compensated for by good performance in other
thiscategory
criteria (as is the case for compensatory MCDA
models,e.g.,utilitytheory).Thisconventionprevents
probability of being classified as ‘‘low to very low decisions about the risk of a particular nanomaterial
risk.’’ C60 showed a reasonable acceptability index being unduly based on one particular criterion (such
(49%) for the low risk category. In terms of making as size versus surface reactivity relationships), as the
risk-aware decisions for C60 and Al NPs, we feel materialmay have otherphysico-chemical character-
that further studies into expanding the potential istics related to size that exhibit a greater impact on
applications of Al NP and C60 (as opposed to CdSe) its toxicity.
are justified.
Itisimportanttopointoutthatinspiteofthehigh Acknowledgment Comments provided by F. Kyle
uncertainty of the above results, this work represents Satterstrom, Jacob Stanley, and David Johnson were very
helpful in the preparation of this manuscript. Permission was
a reasonable starting point for a more thorough
grantedbytheChiefofEngineerstopublishthisinformation.
follow-up analysis. Indeed, more data is required to The studies described and the resulting data presented herein
improve our estimates. Risk estimates based on were obtained from research supported by the Environmental
acceptability indices below a certain threshold (e.g., Quality Technology Program of the US Army Engineer
Research and Development Center (Dr. John Cullinane,
80%) should be viewed with caution. For example,
Technical Director). The views and opinions expressed in
shouldC60bedeemedviableforfurtherresearchand thisarticlearethoseoftheindividualauthorsandnotthoseof
application,additionalmeasurementswillberequired theUSArmyorothersponsororganizations.
to further refine the risk estimates. In spite of its
limitations, the quantified risk values obtained from
References
our simulations are helpful in characterizing the risk
and uncertainty for limited and variable data.
AuffanM,RoseJ,OrsiereT,DeMeoM,AchouakW,Chaneac
C, Joliver J-P, Thill A, Spalla O, Zeyons O, Maison A,
Labille J, Hazeman J-L, Proux O, Briois V, Flank A-M,
Botta A, Wiesner MR, Bottero J-Y (2008) Surface reac-
Concluding remarks
tivity of nano-oxides and biological impacts.
Nanoparticles in the Environment: Implications and
Nanotechnology is a rapidly growing research field Applications, Centro Stefano Fracnscini, Monte Verita,
with an increasing impact on our everyday lives. Ascona
Belton V, Stewart TJ (2002) Multiple criteria decision analy-
Although nanomaterials are used in common con-
sis—an integrated approach. Kluwer Academic
sumerproducts,thelackofinformationabouthuman
Publishers,Dordrecht
health and environmental risks may hamper the full- BiswasP,WuC-Y(2005)Nanoparticlesandtheenvironment.
scale implementation of this technology. In this JAirWasteManagAssoc55:708–746
Borm P, Mu¨ller-Schulte D (2006) Nanoparticles in drug
article, we presented a systematic multi-criteria
delivery and environmental exposure: same size, same
approach that enables nanomaterials to be assigned
risks? Nanomedicine 1(2):235–249. doi:10.2217/17435
into ordered risk classes. Materials assigned to the 889.1.2.235
123