Chemometrics and PAT: What Does it all Mean, J Workman Jr

Tags: PAT, technology, processes, chemometric, requirements, process model, Process Analytical Technologies, chemometrics, statistical methods, application, implementation, real-time data, business methods, data collection and analysis, multivariate approach, plant environments, chemical knowledge, PAT Chemometrics, statistical process control, understanding, cost minimization, control limits, product quality, chemical properties, data collection, Old technology, manufacturing, Standardization, physical test data, Process Analytical, Analytical Technologies, Molecular Spectroscopy, process control, sensor data, information inputs, models, product status, information yields, methods, physical properties, continuous process control, process optimization
Content: Chemometrics and PAT:
What Does it all Mean?
A paradigm shift is required for chemists and engineers to best utilize chemometrics in their processes. This change demands that one not be too fixated upon ideal textbook thermodynamic models but instead continually check these models using real-time data input and chemometric analysis. The author discusses implementation strategies and the benefits that chemometrics can bring to the process environment.
Jerome Workman, Jr.
C hemometrics historically has provided a valuable toolbox for obtaining information from multivariate data. The term "chemometrics" might refer to the capability
duction cycle times and enhance capacity utilization; thereby in the long run, reducing product development time . . . ." (1) The current pharmaceutical manufacturing paradigm is
of computationally analyzing multivariate data in real-time to skewed toward testing to document product quality and
obtain useful information for analytical applications. Re-
rejecting or recalling products of unacceptable quality. For
cently, the term informatics, or the application of computer the PAT paradigm, poor process understanding leads to
and mathematical­statistical techniques to the management poor quality control, delays in drug regulatory approval,
of information, might be a more appropriate term covering disruption of essential drug and vaccine supplies (for exam-
the broader aspects of information management. The utiliza- ple, the recent U.S. flu vaccine production problems), and
tion of this information management is required for efficient lack of available low volume drug supplies. With recent
control of pharmaceutical processes. Whatever terminology is developments, "safe harbor" allowances from U.S. regulato-
used, it is clear that such computational power is key to pro- ry bodies now allow pharmaceutical manufacturers to
viding essential information in real-time from ever increasing implement PAT without the threat of regulatory backlash.
quantities, and variable quality, of data measured from
The main purpose of PAT is for the facilitation of high
process streams. Information management is required during quality medicines produced efficiency, and in adequate sup-
raw material identification, active pharmaceutical ingredients ply. The stakeholders in the successful implementation of
(API) production, product production, and final product
PAT include consumers, manufacturers, and government
packaging. This article briefly discusses the application of in- agencies. The results of PAT implementation are new high
formatics and chemometrics for the implementation of
quality medicines produced economically for improved
process analytical technology (PAT) for a "typical" pharma- health and enhanced economic growth.
ceutical process.
According to many experts a set of metrics is available to
compare the benefit of PAT implementation to benchmark
Goals and Objectives of the PAT Subcommittee and
production of pharmaceutical materials. Many of these met-
Working Groups
rics relate to supply chain management for manufacturing.
PAT hFasobreeCnlideensctriRbeedvaiseiwnvoOlvninlyg."AsylslteRmisgfhortscoRnetinseuoruvsed. ATdhevagonaslstaofrsCupopmlymchuainnieccaotnioomniscsIanrec.to2m00an5ufacture
analysis and control of manufacturing processes based upon equivalent or improved quality goods faster, with increased
real-time measurements, or rapid measurements during pro- reliability, by less expensive means, with reduced inventory.
cessing, of quality and performance attributes of raw and in- For the more business-minded readers, implementing supply
process materials and processes to assure acceptable end
chain management brings the expectation of increases in rev-
product quality at the completion of the process . . . . PAT in- enue, profit margin, cash flow, and competitive position.
volves the use of process analytical chemistry tools, informa-
In a recent report (2) such metrics of concern to pharma-
tion management tools, feedback process control strategies, ceutical manufacturing companies include: Stock Turn, or a
and product and process design and optimization strategies. measure of how quickly a business turns over its supply
PAT used in pharmaceutical applications can improve the ca- (stock) of any product; On Time In Full (OTIF), a percentage
pability and the efficiency of pharmaceutical processes while metric of how often product can be produced on time while
maintaining or improving product quality: improve process meeting production quantity and quality standards; Right The
understanding and help to ensure quality is `built in' or `by
First time (RTF), a metric describing the percentage of prod-
design'. It can reduce the risk of scrap and recalls; reduce pro- uct meeting specification the initial time it was produced;
18 The Role of Spectroscopy in Process Analytical Technologies January 2005
www.spectroscopyonline.com
Process Capability (CpK), as the mean
The analytical measurements must
for a series of production runs minus the lower tolerance limit divided by
provide information to other systems in order for the PAT to be useful. Raw
Process sensors
Controlled process
three sigma for the process; and Overall data from sensors are processed into
Equipment Effectiveness (OEE), the
information used to control the
percentage of time the process equip- process (see Figure 1). As illustrated by
ment is adding value under specific con- this flow diagram, a new or existing
Analytical data
Validated process model
ditions. In the past the more traditional manufacturing efficiencies were not considered a key factor within pharmaceutical companies. Traditional pharma-
process is measured for one or more key parameters by one or more realtime sensors. The sensor information (as raw analytical data) is supplied to a
Chemometric modeling
Chemometric modeling
ceutical manufacturing has followed
computational system, which converts
Information
prescribed analytical and physical tests the raw data into useful process
whereby incoming raw materials (that parameter information (for example,
is, active ingredients and excipients),
spectra, chromatograms, or physical
Figure 1. PAT for continuous process control and
intermediate reaction products, and
sensor data are converted into parame-
improvement.
final products are analyzed using vali- ter information such as chemical or
dated methods. This approach is logical Physical Properties). This information ity between basic measurable (and con-
following the technologies that have
is processed further into deterministic trollable) attributes and product per-
been available in the past, but this strat- process models, which define the
formance.
egy does nothing to optimize capacity, mathematical relationship between
2. Providing methods for assessment
cycle time, and cost reduction. Neither specific manufacturing control param- of traditional test data, which often
does it address an entire host of eco-
eters (for example, solvent concentra- provides marginal real value to process
nomic factors such as inventory control, tion, mixing rate, drying temperature, control, with updated sensor data to
waste, or process fluctuations. Variations pressure, and so forth) and the analyti- provide the best information when and
in process quality over time often will cal information (that is, chemical or where it is needed. Results of a few
result in uncertain and relatively ineffi- physical properties). Once the process select studies have indicated that only a
cient product manufacturing. The appli- model is derived and validated, it is
small fraction of the data collected
cation of PAT promises to allow real-
included in a process information
yields information relative to important
time assessment of product status dur- management system (PIMS) as part of attributes of products and processes (in
ing manufacturing processes applying a broader informatics control system. other words, collecting the right data is
learning and control to improve quality This process model then is used to
essential).
and economic efficiencies.
control the process during production
3. Providing better, faster methods
while the process is measured continu- for assessing and modeling product
PAT and the Pharmaceutical
ously using the installed sensor net-
performance during new product devel-
Manufacturing Process
work. So the analyzed sensor data pro- opment stages, with a purpose to enable
PAT provides a basis for monitoring the vides information inputs for control of more rapid deployment of products
manufacturing process, determining
the manufacturing processes in real- with an understanding of chemistry,
control parameters for that process, and time. Both the inferential model and versus structure, versus function, versus
modeling these parameters to produce the process are improved as new
process methods. This approach also
optimized production. If during a man- process information is gathered by the can be used to assess the attributes that
ufacturing process the measured pa-
sensors and incorporated into both
are most important during invention of
rameters indicate a process disruption sensor calibration models and process new products and enhancement of or the prodFuoctrisCmlioevnintgRtoewvairedwnoOn-nly. AclolnRtriogl hmtosdReles.served. Advanstar CopmromduuctnpiecrafotrimonansceInfeca.tu2r0es0. 5
conformance, corrective actions can be
This sensor data mining approach to 4. Looking for improved metrical
initiated. The acronym PAT includes the modeling, using chemometrics and pat- methods for assessing pharmaceutical,
term analytical, but PAT itself indicates tern recognition techniques, provides sensory and tactile product perform-
much more. In strictest terms, the ana- insight into the relationships between ance from a structure­compositional
lytical portion of PAT provides the "eyes the physical and chemical properties of standpoint. This includes moisture
and ears" of process control, but the
products and their intermediate or final analysis, hardness, texture, odor analy-
brains, hands, and feet involve opera- performance characteristics. After col- sis, appearance analysis, and the like.
tors, chemical and pharmaceutical
lecting the appropriate sensor data from 5. Developing improved methods for
process engineers, synthetic chemists, a process and applying essential com- understanding the processes involved
statisticians, mechanical engineers,
putational tools, multiple objectives can in specific core technology areas such
quality assurance professionals, and in- be accomplished:
as drying, mixing, blending, crystalliza-
formatics and control experts, among
1. Relating chemical, physical, and
tion onset, fermentation, synthesis,
others.
image (structural) data to define causal- tabletization and other technologies
January 2005 The Role of Spectroscopy in Process Analytical Technologies 19
Chemometrics and PAT
related to core pharmaceutical business of preprocessing, calibration, and diag- means to the real purpose of gaining an
competencies.
nostics; and rigorous prediction valida- understanding of the fundamental sci-
PAT-associated technology provides tion. It should be pointed out that the ence involved. Quality data from appro-
an opportunity for organizations to
initial DOE determines the extent to
priate DOE combined with understand-
take a managed approach to under-
which chemometric models can be
ing of the chemical and physical phe-
standing the science relative to products applied. If the DOE accurately describes nomena are essential ingredients. In
and processes important to their organ- the actual use conditions then the
addition, the limitations of statistical
izations. The expected outcomes from application of the model derived from methods must be accepted for success-
this work would include improved
this experiment will apply for these
ful long-term implementation of
product quality and production effi-
specified use conditions. Conversely, if chemometric models to the under-
ciency, but also the potential for devel- the DOE is "highly contrived" and lim- standing and efficient control of manu-
opment of proprietary mathematical ited the models derived from it cannot facturing processes.
tools, business methods, processes,
be applied with certainty under real
Some specific standards activities rel-
products, devices, and composition of world use conditions.
evant to PAT involve the ASTM
matter (that is, new chemical entities
The application of chemometrics in International E55 Main Committee
[NCEs] and dosage forms of these new PAT requires the application of proven tasked with producing standard termi-
molecules). These practices and materi- mathematical and statistical methods nology, guidelines and practices for PAT
als should result in competitive advan- and a basic understanding of chemical to be used as international consensus
tage from new intellectual property
phenomena under study. This implies methods for process measurements and
(IP), including patents and trade
that the model makers should be strong control. One of the group's first formal
secrets. Such IP benefits include: math- in chemical knowledge with a basic
activities is referred to as E2363-04,
ematical techniques and informational understanding of sound statistical
"Standard Terminology Relating to
approaches, control strategies, identifi- methods, rather than formally trained Process Analytical Technology in the
cation of essential tests to provide qual- statisticians with some slight familiarity Pharmaceutical Industry" (5). This doc-
ity data, and improved understanding with the chemistry and pharmacy. The ument in draft contains the terminolo-
of process and product attributes.
process modeling will be deterministic gy and definitions of concepts impor-
and will provide for the derivation of tant for precise communication of PAT
Chemometrics as an Essential
the state of the process -- thus the
nomenclature. Another useful docu-
Ingredient of PAT
focus is chemical understanding, not ment applicable for chemometric meth-
Chemometrics generally can be de-
statistical tool making. Once a model is ods is ASTM E1655-00, "Standard
scribed as the application of mathemat- implemented it must be monitored
Practices for Infrared Quantitative
ical and statistical methods to improve during the process using actual real-
Analysis." This document is useful
chemical measurement processes by ex- time measurements of the process. With across multiple spectroscopic methods,
traction of useful information from
the monitoring system in place, the
delineating instrumentation require-
chemical anFdoprhyCsliicealnmteRaseuvreimewentOnly. AprllocResigs ihs tmsaRnueaslleyrovreadu.toAmdavtiacanllsytar Commenmts,ucnaliicbarattiioonnms aInthce.m2a0ti0cs5, statis- data (3). Chemometrics is used for mul- controlled with real-time feedback pro- tics, pre- and postprocessing methods,
tivariate data collection and analysis
vided to individuals having a detailed outlier statistics, calibration and valida-
protocols, calibration, process model- understanding of the chemical and
tion protocols, troubleshooting guide-
ing, pattern recognition and classifica- physical phenomena involved in the
lines, quality statistics, procedures for
tion, signal correction and compression, manufacturing process as well as an
updating models, terminology, and a
and statistical process control. Standard understanding of the limitations of the questionnaire to check compliance with
terminology, practices, and methods are statistics to determine ultimate causal the standard (6). Other standard meth-
required for "verification" and "valida- relationships.
ods for the application of near infrared
tion," at least for those that will seri-
The results of expert analysis involve spectroscopy in particular are under
ously implement and evaluate such
improved understanding of products draft development at the time of this
methods. Developers have more free- and processes based upon a priori
article.
dom to explore new techniques, but at appreciation of the basic science associ- The technical advantages of using
the end of the day they must pass rigid ated with product attributes. Statistics chemometrics include: speed in obtain-
standard "validation" testing.
alone does not demonstrate cause and ing real-time information from data;
The chemometrics working group of effect. The technical team performing the capability of extracting high quality
PAT has made recommendations for the this work must explore the underlying information from less resolved data;
use of chemometrics in PAT implemen- principles causing variation within the and the capability of clear information
tation (4). In these recommendations, data both to improve the computation- resolution and discrimination power
the key aspects important to the suc- al approaches to understanding the
even when applied to second, third, and
cessful application of chemometrics
data, and to assess the fundamental
possibly higher-order data. Another key
involve: scientifically valid design of
causal process information. For this
benefit includes the potential for pro-
experiments (DOE); proper application work statistics are useful, but only as a viding methodology for cloning sensors
20 The Role of Spectroscopy in Process Analytical Technologies January 2005
www.spectroscopyonline.com
Chemometrics and PAT
-- for making one sensor take data
A paradigm shift is required for
such as recognition for accomplish-
"precisely" as another sensor. Additional process specialists in order to best uti- ment, demonstrated process improve-
technical benefits include: diagnostics lize chemometrics. This paradigm
ment, no risk, convenience, and eco-
for the integrity and probability that
change demands that one not be too
nomical choices. In other words, the
the information derived from sensor
fixated upon ideal textbook thermody- risk/reward ratio must be near zero.
data is accurate; improved measure-
namic models which always occur pre- The company has a separate list of
ment quality; and, eventually, improved cisely "as advertised" under most condi- requirements, such as improved process
knowledge of existing processes.
tions. A revised way of thinking
performance, increased profits, mainte-
economic benefits of chemometrics requires a reality check on these ideal nance or improvements in quality, con-
include: relatively low capital require- states involving real-time data input
venience, low cost implementation, and
ments; safer plant and process opera- from many inexpensive measurements, low risk. Thus the [rewards/(risk +
tions through real-time monitoring
followed by chemometric analysis. The cost)] ratio must be a very large num-
and prevention of potentially danger- revised paradigm looks at all the data ber. Chemometrics supplies a perfect fit
ous process upsets; assurance that
from a multivariate approach, whereas to these requirements by providing the
processes and plant environments are more traditional thinking assumes that expertise (time and talent) into the
in compliance with environmental reg- ideal thermodynamic models combined resource equation. Chemometrics is
ulations; and an increase in process
with the operator's powers of observa- applied at a minimum cost using data
plant operability through timely
tion and a dash of intuition and leg-
analysis techniques requiring compara-
adjustments in processes possible using erdemain are the key data processors. tively small investments in sensor and
real-time data. Other benefits are:
Thus for the new paradigm, nothing is computer time, and demonstrating a
improved product quality through
assumed to be ideal -- it follows the
potential benefit in understanding a
maintenance of tighter control limits; principle that "one designed experiment process providing priceless information
minimization of waste products
is worth a thousand educated opin-
for improving it. Risk is minimized due
through process optimization; product ions." And so consider the information to the flow of real-time information.
production cost minimization through content of a thousand well-measured
A check-list for making chemometric
tighter target limits and more accurate results versus that derived from a pre- sensors work includes: test underlying
production scheduling; optimization of sumed textbook process model with a assumptions continually and thorough-
production capacity resulting from
few selected measurements.
ly at the lowest possible cost; prepare
increased process operability and con-
Typically, the process decisions are in multiple alternatives solutions; commit
tinuous product quality verification. the domain of the chemical engineer, to implementing technology but not to
Chemometric-based real-time meas- plant manager, and quality group. Their one use or application of a technology
urements eliminate the greatest chal- process decisions are based upon
-- look for multiple technologies and
lenges to 100% compliance and analyt- process modeling, temperature, pres- multiple uses; manage time and cost
ical accuracy for measuring a process sure, flow, and mass balance, which
commitment issues; and avoid overload
-- namely, sampling error. In addition, generally are considered adequate using of staff (for example, two substantial
well-managed PAT will provide insight traditional evaluation criteria. Decisions projects each is optimum). Additionally, into new inFteollrecCtuliael nprtoRpeervtyiefowr cOomn-ly. Aalrle Rmiagdhe tins tRhee spelarnvteandd. Athdrovuagnhsvtaarir- Cotmhomse umnanicaagitnigotnhsis Iinntce.rn2a0l 0pr5ocess
petitive advantage opportunities.
ous engineering groups. These decisions should determine if there is an internal
On the downside, anyone with a
are based upon past experience and
customer market for this technology or
computer can generate complex
current academic training. The reason approach; if the technology can be
numerical solutions that can be impos- changes in the decision-making modus delivered reliably and cost-effectively;
sible to interpret, and complex mathe- operandi are so slow, and why many
and if small exploratory forays into less
matical solutions can leave plenty of
resist changes including chemometric- challenging opportunities can be made,
room for misinterpretation unless strict based sensors, is precisely due to
learning and adapting as progress is
guidelines (as established through stan- resource deficiencies -- in time, talent, made. The entire scheme is codified and
dards) are followed. Also, chemomet- attention, motivation, and economic
diffused continually throughout the
rics requires a change in one's
incentive. But what happens when one organization, including risks, potential
approach to problem solving from uni- demands 100% compliance with no
benefits, and proposed solutions. This
variate to multivariate thinking, and batches out of specification? Under
procedure results in a learning and
effective use requires a paradigm shift such new constraints something must adaptive organization.
and a process model understanding
be done to improve operating proce-
closer to reality (honesty). Finally, most dures -- thus PAT is needed when the The Challenge of Transitioning
"best practices" still need to be collect- quality and economic status quo is
to PAT
ed and codified into useful standards challenged.
technological change is difficult for sev-
-- requiring much cooperative effort
The process engineer and manufac- eral reasons. New technologies usually
between practitioners, theoreticians,
turing personnel require motivators if are inferior to present state of the art;
and regulatory groups.
their behavior patterns are to change, today's technology leaders dismiss the
22 The Role of Spectroscopy in Process Analytical Technologies January 2005
www.spectroscopyonline.com
new technology as they are generally
by an initial announcement of a data- requirements for quality. With all its
unfamiliar with it; new technology can mining project specifying the types of sophistication in chemical, mechanical,
intimidate the uninitiated after take-off; data requested and the purpose for the physical, and computational technolo-
past success creates the seeds of compla- request. This will require no more
gies one must not forget that for the
cency due to arrogance; this compe-
interaction from the individuals or
implementation of PAT, R.B.
tency traps itself in the status quo; and groups supplying data other than the Woodward's famous quote is suitable:
for competitive leadership, the compe- release of the data in a readable spread- ". . .One's ideas, however beautiful, logi-
tent must seek to replace themselves
sheet format and a brief explanation of cal, elegant, imaginative . . . are simply
with new competencies or become less the data parameters. An electronic form without value unless they are actually
competitive and eventually face eco-
could be included with the data to sim- applicable to the one physical environ-
nomic extinction. Old technology in- plify the transfer process. Additional
ment we have; in short, they are only
sists on improved execution of the "less testing might be required in some stud- good if they work." Furthermore, they
optimal" thing, not on an emphasis on ies to verify causality. In the early or
are usable only if they can be validated
doing the "most optimal" thing.
later stages new sensors can be installed by stringent regulatory standards,
During the implementation of PAT, to provide better information content which follow from rigorous science and
one would be helped by careful man- for assessing important processing
engineering discipline.
agement of adaptation and change. This parameters.
can be assisted by noting that one can
The business objective of PAT imple- References
gather fast and cheap data to start with mentation is to develop the technology 1. A.S. Hussain, "The Subcommittee on
and assess the quality of this data before to perform data mining on chemical
Process Analytical Technologies (PAT):
proceeding. Problems at this stage gen- and physical test data of a manufactur-
Overview and Objectives," FDA's Advi-
erally are easier and cheaper to solve. ing process to derive understanding of
sory Committee for Pharmaceutical Sci-
Remember that data need to be man- fundamental scientific principles rela-
ence, Panel Meeting in Gaithersburg,
aged to provide more information; it is tive to measurement tests, products,
MD, February 25, 2002.
this information (not raw data) that
and processes for competitive advan- 2. I. Clegg, Committee E55 on Process Ana-
provides potential learning for
tage strategies.
lytical Technology, Standardization
improved processes. The size of your
PAT implementation milestones can
News, pp. 28­31, May 2004.
information "pile" indicates the learn- include the set-up and documentation 3. J. Workman, P. Mobley, B. Kowalski, and
ing potential for future success. Data
of appropriate sensor, software tools,
R. Bro, Appl. Spectrosc. Revs. 31,
alone have no value for learning until it methods, approaches, and applications
73­124 (1996).
is converted to relevant information
for data collection and modeling of
4. Chemometric Working Group Recom-
supplied to decision makers.
processes. Additional milestones include
mendations, The Subcommittee on
It is essential to fix your variables
the development and documentation of
Process Analytical Technologies (PAT),
during experimentation, thus the use of proprietary calibration and process
FDA's Advisory Committee for Pharma-
standardized calibration and modeling modeling techniques, and the applica-
ceutical Science, Panel Meeting in
algorithms is essential for making sense tion of these techniques for solving a
Gaithersburg, MD, February 25, 2002.
out of dataFcoolrleCctleidenovteRr aevsiinegwle oOr nly. AtelclhRniigcahlltyschRaellesnegrivngedan.dAidmvpaonrtsatnat r Co5m. TmheuAmneicriacatnioSoncsietIynfocr.T2es0ti0ng5and Ma-
multiple experiments. In addition, hav- problem, or series of important prob-
terials (ASTM) International Main Com-
ing standard samples or test materials lems. The third milestone would involve
mittee E55 on Process Analytical Tech-
is essential for sensor monitoring and the presentation of the findings from
nology, ASTM International, West
calibration transfer. Sensor systems
the various studies completed and rec-
Conshohocken, PA 19428-2959 USA.
should be established with standard
ommendations for implementation and 6. The American Society for Testing and Mate-
"outlier" detection methods to deter- use in process control. A fourth might
rials (ASTM) International Main Commit-
mine whether sensors, sampling sys- involve the transfer of this technology
tee E13 on Molecular Spectroscopy and
tems, or processes are performing out- to appropriate process groups for con-
Chromatography, ASTM International,
side of expected ranges. Moreover,
tinued development and use of the con-
West Conshohocken, PA 19428-2959
standard analyzer functionality tests
cepts discovered and real-time manual
USA, Vol. 03.06 , 534­561 (2004).
capable of diagnosing instrument mal- or automated production process con-
function are important. Finally, stan- trol. Typically pilot plant operations
Jerome Workman Jr. serves on
dard calibration and validation proto- could be used for feasibility and later
Spectroscopy's Editorial Advisory Board and
cols should be implemented which are limited production of final product
is co-editor of the magazine's
based upon sound principles of experi- tested. Later validation of the method "Chemometrics in Spectroscopy" column. He
mental design.
for product release could be undertaken is director of research, technology, and appli-
Potential difficulties are anticipated or the PAT could be used to make
cations development for the Molecular
relative to the release of proprietary
adjustments to the process to ensure
Spectroscopy & Microanalysis division of
data from certain groups or individuals that 100% of production material
Thermo Electron Corp. He can be reached
to the PAT Team. This can be overcome meets the validated analytical test
at: [email protected]
January 2005 The Role of Spectroscopy in Process Analytical Technologies 23

J Workman Jr

File: chemometrics-and-pat-what-does-it-all-mean.pdf
Title: SPPAT105_18-23
Author: J Workman Jr
Author: d Himes
Published: Fri Jan 14 16:11:00 2005
Pages: 5
File size: 0.15 Mb


, pages, 0 Mb

Confronting disaster losses, 7 pages, 0.13 Mb

Becoming a teacher, 17 pages, 0.05 Mb

THE BIG IDEA, 7 pages, 0.25 Mb

THE POET IN VAN MORRISON, 14 pages, 0.1 Mb
Copyright © 2018 doc.uments.com