Research Policy 35 (2006) 1–23

Exploration and exploitation in innovation systems:

The case of pharmaceutical biotechnology
Victor Gilsing a,∗ , Bart Nooteboom b
a ECIS, Eindhoven University of Technology, TEMA 0.32, P.O. Box 513, 5600 MB Eindhoven, The Netherlands
b Tilburg University, The Netherlands

Received 24 April 2004; accepted 11 June 2005

Available online 19 December 2005

Abstract

This paper develops a theoretical framework for an explanation of how exploitation and exploration build on each other, in
a ‘cycle of discovery’, developed in earlier research. The framework is tested empirically, in the sense of seeing whether it
can help to reconstruct and understand the emergence of the pharmaceutical industry. One of the conclusions is that whereas
recent literature stresses the idiosyncratic nature of the biotechnological revolution, our analysis seems to reveal that this does
not seem to be as unique as suggested. From this, we conclude that the theoretical framework we propose, serves its purpose of
explanation. But there are also some lessons for improving it.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Exploration; Exploitation; Pharmaceutical biotechnology; Sectoral systems

1. Introduction ‘dominant designs’ and mindsets regarding product,

Many studies have analysed the effects of a wide
range of institutions on innovation (Lundvall, 1992;
Nelson, 1993; Edquist, 1997; Whitley, 1999; Mowery
and Nelson, 1999; Malerba, 2002, 2004). Some insti-
tutions are industry-specific, such as systems of pro-
duction, organization, distribution, supply, technical
standards, training, and so on. Such institutions may
be associated with ‘industry recipes’ (Spender, 1989),
∗ Corresponding author. Tel.: +31 40 247 4435;
fax: +31 40 246 8054.
E-mail addresses: v.a.gilsing@tm.tue.nl (V. Gilsing),
b.nooteboom@uvt.nl (B. Nooteboom).
market and organization, in systems of ‘exploitation’
(March, 1991). Broader national institutions, beyond
industries, include government, legal systems (includ-
ing property rights), infrastructure, general training and
education, labour conditions, financial systems (includ-
ing venture capital), and the like. These have had a
clear influence, e.g. on the biotechnology revolution,
as documented in a range of studies (e.g. Gambardella,
1995; Orsenigo et al., 1998, 2001; Henderson et al.,
1999; Jungmittag et al., 2000; Pisano, 2002; Gassman
et al., 2004; McKelvey and Orsenigo, 2004). We fully
acknowledge their importance, but in the present paper,
we do not aim to repeat or extend the study of their
effects. We are interested, rather, in the basic ‘system

0048-7333/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.respol.2005.06.007
2 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
logic’ of how, in a new round of radical innovation,
sector-specific institutions such as dominant designs,
industry recipes, and mindsets change. In other words,
how exploration builds on but also shifts existing sys-
tems of exploitation. This system logic is based on a
theory of learning, in the sense of discovery, devel-
oped in earlier work, which entails a dialectic process
of exploitation and exploration (Nooteboom, 2000),
intended to explain how radical innovation and struc-
tural change arise endogenously. In this paper, we dis-
cuss this logic at a sectoral level, to understand the
transition from exploitation to exploration and vice
versa. On the basis of this, we analyze the effects of
such sectoral dynamics on organizational forms, in par-
ticular networks of firms.
We apply this framework to the pharmaceutical
industry to see whether it can help reconstruct and
understand the biotechnological revolution. Recent
studies on the pharmaceutical industry seem to agree
on the fact that there have been, apart from firm-
specific and national aspects, fundamental changes
at the sectoral level (Arora and Gambardella, 1994;
Gambardella, 1995; Orsenigo et al., 1998; Henderson
et al., 1999; Jungmittag et al., 2000; Pisano, 2002;
Brusoni and Geuna, 2003; Santos, 2003; Gassman
et al., 2004; McKelvey and Orsenigo, 2004). This
literature has described in great detail how the advent
of molecular biology and genetic engineering yielded a
profound transformation of the pharmaceutical indus-
try and induced a new division of labour that required a
new organizational form made up of networks of scien-
tists, specialised new entrants and large pharma firms.
The claim that we submit here is that our system logic
helps to understand these sectoral dynamics. In this
way, we aim to go beyond some of the more descriptive
accounts as present in the recent literature and aim to
contribute to an evolutionary theory at a sectoral level,
which holds across different technologies or industries
(Nelson, 1994). Apart from the sectoral developments
that have taken place on a worldwide scale, we analyze
how such sectoral dynamics have settled in network
structures in the Netherlands. Although the literature
abounds with claims and empirical evidence on the
role of networks in this industry, a more in-depth
understanding of how they are structured and of their
inner functioning is still limited. Analyzing such
networks in more detail echoes the claim by Powell
et al. (1996) that networks are a more useful concept
for analyzing innovation in pharmaceuticals than
firms.
This paper looks at issues of both competence
(learning, innovation) and governance (management of
relational risk). We propose that it is the combination
of the two that yields a more complete understand-
ing of interfirm networks (Dosi and Marengo, 2000;
Nooteboom, 2000). We identify two main kinds of rela-
tional risk: the hold-up risk familiar from transaction
cost economics, as a result-specific investments, and
risk of spillover of strategically sensitive knowledge to
competitors. Specific investments may occur in other
ways than recognized in transaction cost economics, in
building up mutual understanding and relation-specific
trust, which are both important, in particular, under
the uncertainty of exploration, with the development
of new knowledge. In contrast with transaction cost
economics, we claim the viability, and indeed indis-
pensability, of trust that, within limits, goes beyond
calculative self-interest. The result is that we employ
a range of instruments of governance that include
coercion by hierarchy or contracts, incentives from
dependence on unique partner value, hostages, and rep-
utation, and trust based on ethics, shared values of
conduct, and relation-specific empathy, identification
and routinization of conduct (for an integrated account,
see Nooteboom, 2004). The paper proceeds as follows.
Section 2 discusses the theoretical framework, a ‘cycle
of discovery’, which serves as a logic of how explo-
ration and exploitation are related and build on each
other. In Sections 3 and 4, for an empirical test of
our claims, we describe and analyse the transformation
process of the emerging pharmaceutical biotechnol-
ogy industry (Section 3), and discuss its organizational
implications in the Netherlands (Section 4). In Section
5, we compare our empirical findings with the claims
based on our theoretical framework and we draw some
conclusions.
2. Exploration and exploitation: a cycle of
discovery
Given the aim of this paper to provide a deeper
understanding of the transition from exploitation to
exploration and vice versa, we need a theory that
explains it. For that we use a ‘cycle of discovery’
(Nooteboom, 2000) that describes and explains how
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Fig. 1. Cycle of discovery (Nooteboom, 2000).
exploration and exploitation are mutually related and
build on each other.
Exploitation, i.e. the efficient employment of current
assets and capabilities, is needed to survive in the short
term. Exploration, i.e. the development of novel capa-
bilities, is needed to survive in the long term. Thus, to
survive in the short and long term, firms must combine
the two, somehow. That is a paradoxical task. A key
question now is how exploitation and exploration build
on each other. For this, Nooteboom (2000) proposed
a heuristic ‘cycle of discovery’, with several stages,
illustrated in Fig. 1. A key feature of this cycle is an
alternation of variety of content and variety of context.
As described in the standard innovation literature, a
variety of alternative forms of a novelty is reduced,
by a process of selection, in convergence on domi-
nant designs, in technology and organization, in a drive
towards exploitation. Next, and this part of the cycle is
new, for exploration practices must be confronted with
novel conditions and opportunities, in a variety of new
contexts. The cycle is discussed in more detail below.
2.1. Exploitation
Exploitation starts when variety of content (of a
concept, technology, product or practice) that emerges
from exploration is reduced, in consolidation, into a
dominant design, as suggested in the innovation lit-
erature. As a result of reduced uncertainty, demand
3
increases, and new producers jump on the bandwagon.
In manufacturing industries, the focus shifts from a
focus on product innovations to process innovations
(Abernathy and Utterback, 1978). The new technol-
ogy/product/market combination develops into a dom-
inant design or ‘dominant logic’ (Bettis and Prahalad,
1995) of organization, including network structure and
‘industry recipes’ (Spender, 1989), with pressures to
conform, in ‘organizational isomorphism’ (DiMaggio
and Powell, 1983). Competition shifts to efficient pro-
duction and distribution and new entrants exert further
pressure on price. Therefore, for the sake of efficient
production, an increase of scale, a division of labour
and associated specializations emerge. As a result,
variety decreases and the focus shifts to ‘single-loop’
learning between firms (Argyris and Schön, 1978),
yielding only minor product adaptations and process
improvements. As a consequence, there is an increase
in specialization that entails more specific knowledge
on a narrower range of issues.1
1 This clearly connects with Callon’s notion (2002) of a ‘consol-
idated configuration’. Such a consolidated network is characterized
by a strong similarity of the involved actors that face limited uncer-
tainty as the possible states of the world are known ex ante and
can therefore be anticipated. As a consequence, they develop simi-
lar research and innovation programs leading to a further narrowing
of technological options and a further increase in the consolidation
of these ‘aligned’ networks. The natural outcome of this process is
formed by incremental innovations.
4 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
2.1.1. Implications for innovation networks
When novelty gets consolidated in dominant
designs, knowledge gets more codified, which enables
a more rapid diffusion. A need and a basis arise for more
formal governance (Nooteboom, 2004). There is a need
because relations become more distant and therefore
less personalised, and codification of knowledge yields
a greater risk of spillover. There is a basis because
there is more information about options, outcomes and
tasks for constructing contractual arrangements, and
codification of knowledge yields better opportunities
for contract specification and monitoring. A reputation
mechanism has developed. Trust shifts from relation-
specific to institutions-based trust (Nooteboom,
2002), as the appropriate institutions arise, in the
form of technical and behavioural standards, rep-
utation mechanisms, and legal and organizational
forms.
Stringent technical requirements and opportunities
of more codified information, reduced uncertainty, spe-
cialization, and more formal governance are best facil-
itated by ties that are strong in some respects and
weak in others. Ties are strong in terms of high and
often specific investments in the organization of pro-
duction and distribution, and in durability, to ensure a
stable, efficient division of labour, and a pay-back to
the investments. The drive for efficiency requires the
elimination of redundant relations. This pushes for a
less dense structure of the network. Increased codifi-
cation of knowledge and the emergence of standards
further diffusion without the need for relation-specific
investments of mutual understanding. There is less
need for variability and variety for exploring novel
combinations. Increased competitive pressure narrows
the potential for trust and creates a need for contracts.
At the same time, reduced uncertainty and more cod-
ified, diffused knowledge on a narrow range of issues
enable the specification of contracts and the monitor-
ing of compliance. So, whereas ties show strength in
terms of duration, ties show generally lower strength
in terms of lower frequency of contacts for knowledge
exchange, less mutual openness and a narrower range
of issues included in ties.
The interesting question next is how one gets away
from the dominant designs in technology and from
prevailing industry recipes regarding organization, for
a next round of (radical) innovation. In the litera-
ture on innovation and evolution, this question has
been neglected, relative to the question how nov-
elty settles down into dominant designs and gets
diffused.
For exploration that yet maintains such exploitation,
one next needs to open up to a new variety of contexts
of application, in generalisation. These novel contexts
of application may be sought voluntarily, in an expan-
sion of activity. In economics, there is a pressure to
extend the market as growth in the original market
stagnates. While firms may actively search for novel
contexts of application, new conditions of market and
technology may also be imposed from outside one’s
familiar niche. An illustration of this, in the develop-
ment of the multimedia industry, is that publishers were
first not very pro-active in developing the potential and
use of Internet, and went along in digitalization and
electronic distribution of text only later, for fear of los-
ing their position (Gilsing, 2005).
A novel context, either sought outside or imposed
from outside, is needed for three reasons (Nooteboom,
2000). The first is that established capabilities arose
and consolidated in a given niche, and therefore per-
form well there, and are taken for granted, so that new
conditions are needed to gain new insights in limits
of validity. The second reason is to build motivation
for change, resulting from such limits, in the novel
context. The third reason is to yield insight into poten-
tial novel content of practice, for which inspiration is
found in the novel context. Thus, the move of gen-
eralization, into a novel context, yields the basis for
exploration.
2.2. Exploration
First, to maintain exploitation as much as possi-
ble, there is an attempt to make minor adjustments to
established practice, in what is called differentiation.
Insights for differentiation may come from previous
experience, in novel selections from familiar reper-
toires, which are retrieved in an attempt to improve fit in
the novel context. Insights may also come from local
knowledge embedded within the novel context. Key
is that these insights easily fit into existing practice.
As a consequence, innovation is incremental, com-
plexity stays limited and existing network structure
suffices.
However, such incremental adaptations may turn to
be insufficient to fit to the novel context. When differ-
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
entiation is not sufficient, or if the novel context indi-
cates novel opportunities, more profound changes are
needed, in reciprocation: experiments are conducted
with novel elements adopted from the novel context,
which seem to be successful where familiar practice
fails, in hybrids of established and new elements. It
may turn out that in some respects these new elements
perform better in the novel context, so that one tries to
incorporate such elements in the established practice,
or adopts local practices while incorporating elements
from the established practice. This allows for exper-
imentation with new elements, to test their potential,
while existing basic design principles are increasingly
being questioned. In other words, the phase of recip-
rocation forms the ‘topple point’ between exploitation
and exploration. On the one hand, it still allows for
ongoing exploitation, albeit in new forms, whereas
on the other hand it provides insights concerning the
potential of novel elements, and concerning the con-
straints imposed by existing designs on the realization
of that potential. Once this hybrid practice starts to
yield diminishing returns, inconsistencies, and added
complexity, this provides an incentive and insight for
more radical architectural change, in novel structures
of old and new elements. This is the stage of ‘explo-
ration of novel combinations’. The existing network
structure no longer suffices and needs restructuring in
order to open up to new variety of content. This stage
of exploration entails a radical reconfiguration of old
systems of exploitation that are no longer consistent
with emerging novelty.2
In this scheme, one can recognise the principles
of evolutionary thinking: consolidation entails selec-
tion among novelty, generalization entails transmis-
sion, and differentiation, reciprocation and exploration
generate new variety of forms. However, the latter
process of variety generation has been neglected by
evolutionary economics and innovation theory, which
have taken the generation of novelty as random (by
2 This clearly connects with Callon’s notion (2002) of an ‘emer-
gent configuration’. Such an emergent network forms a break away
from existing competences in consolidated networks, resulting in
high uncertainty as the possible states of the world are unknown
and only to be understood through trial and error. As a consequence,
research and innovation ‘programs’ are in a constant state of flux and
develop along the way through a process of cooperation and mutual
learning. In analogy with our logic proposed here, such emergent
networks may give rise to consolidated networks and vice versa.
5
analogy to evolution in biology) or not amenable to
explanation.
2.2.1. Implications for innovation networks
Concerning forms and instruments of governance,
we argue that formal contractual safeguards are prob-
lematic in innovation, especially in the stages of more
radical change (Nooteboom, 2004). Emerging novelty
is by definition uncertain, which makes it difficult to
specify outcomes upfront. In addition, a contract needs
to be monitored, which to a large extent is determined
by the nature of the knowledge base: when knowledge
is tacit and subject to radical change, monitoring may
be very difficult. In addition, there is a relational argu-
ment that detailed contracts aimed at the prevention
of opportunism are a signal of distrust (Nooteboom,
2000). This is especially the case when relations are
(mainly) informal and coordinated by mechanisms
such as social norms and reputation, putting a strong
limit on the use of contracts. As a consequence, the
more radically new exploration/innovation is, the more
one can only use more informal instruments such as
relation-specific trust, and reputation by word of mouth
or early results, such as first prototypes, within the
developing innovation community. It is known from
studies of trust that it is stimulated by the condition
that one needs each other, and there are no alternatives.
Then one will simply have to make it work, in mutual
give and take. Trust is further engendered by mutual
respect among professionals struggling with shared
problems. An informal safeguard against free-riding
may be that in order to keep up with new knowledge
development, all partners have to do their part, in give
and take, to keep up with the development of their
absorptive capacity, and shared tacit knowledge, to ben-
efit from what partners develop. An informal safeguard
against hold-up on the basis of specific investments is
that no one knows yet which investments will turn out
to be generic and which specific. An informal safe-
guard against the use of confidential information as a
hostage is that it is not certain what knowledge could
be used as a hostage. In radical innovation, knowledge
may soon become obsolete, and then is not functional
as a hostage.
In this situation, risks and opportunities for oppor-
tunism are uncertain, and this uncertainty is shared by
all. Opportunism and poaching are restrained by an
unknown risk of retaliation, and by the reputational
6 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23

risk of destroying options for future collaboration. This Table 1

risk also is not known, since one does not have a clue Key characteristics of exploration and exploitation
what options for what kinds of relations, and with Exploration Exploitation
whom, may be at stake. Furthermore, since it is not Competence
clear yet what designs will ultimately be accepted in Radical innovation Incremental innovation
markets, and there is no manifest demand yet, com- Technology oriented Product and process oriented
Experimentation with novel Experimentation in
petition is hardly an issue. As a result there is mutual combinations organization
dependence in unknown risks, concerning technology, Tacit knowledge Codified knowledge
products, markets, and options for future relations. All
Governance
one can do, and can afford to do, is gamble on trustwor- Spin-offs, new entrants Entrance by incumbents
thiness, which may be reciprocal and self-sustaining. Loose alliances Formal alliances, acquisitions
This informal, unspecified governance requires close Limited use of contracts Contracts
and densely embedded interaction, which is needed Relation-based trust Institution-based trust
also for the lack of codified knowledge. So, the net- Networks
work structure needs to be dense to yield reputation Dense, open networks Non-dense, more exclusive
mechanisms, needed in view of the limited feasibil- networks
Informal, flexible ties Formalisation
ity of contractual control. Moreover, density enables Limited size, high entry, Stabilisation
to utilize third parties to aid judgement of the meaning and exit
and value of knowledge (triangulation), and to aid in its Locally embedded Delocalised
absorption. The cost of redundant relations is both lim- Strength of ties
ited, in view of limited size of relation-specific invest- High frequency of interaction Low frequency of interaction
ments, and of limited relevance, since in exploration Short duration Long duration
competition is less on price than on feasibility and fast High(er) openness Limited openness
prototyping. Transitional process
With regard to tie strength, in exploration uncer- Divergence in knowledge Convergence in knowledge
tainty is wide ranging, covering many possible issues and organization and organization
Variety through break-up of Selection by the institutional
of technology, market and organization, and as a result existing networks and environment
ties tend to be strong in the range of issues involved new relations to outsiders
as well as in trust and openness, while generally weak
in formal control. So, trust is needed, next to repu-
tation mechanisms, due to the limited feasibility of 3. Pharmaceutical biotechnology
contractual control. Moreover, to build mutual under-
standing one may require relation-specific investments, Biotechnology as such is not an industry, but refers
which require sufficient frequency of interaction and/or more to a set of technologies that profoundly affect
duration, to make such investments worthwhile. How- existing industries such as agriculture, food-processing
ever, duration need not be very long, since in view of and human health (Pisano, 2002). In this section, we
fast knowledge change specific investments in mutual focus on the impact of the biotechnological revolution
understanding have a short economic life. Duration on the global pharmaceutical industry, and how it set-
should not be too long for two reasons. First, it should tled in network structures in the Netherlands.3 Since
not inhibit fast re-configuration of ties, to enable explo- the biotechnological revolution in the pharmaceutical
ration of novel combinations. Second, long duration industry has been largely global in nature (Henderson,
may yield too much identification, in reduction of cog-
nitive distance, killing learning potential (Nooteboom,
3 We thank Bo Carlsson (Case Western Reserve University, Cleve-
2000; Wuyts et al., in press). Specific investments are
recouped, mostly, on the basis of frequent interaction, land, OH), Patrick Cohendet (University Louis Pasteur, Strassbourg),
Christien Enzing (TNO-STB, Delft, the Netherlands) and Felix Jan-
which is needed also in view of the wide range of issues szen (Rotterdam School of Management, Erasmus University) for
involved in ties, and for building relation-specific trust their valuable comments on an earlier version of this section. Of
(Table 1). course, any error remains entirely our responsibility.
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
1994; Jungmittag et al., 2000), we cannot ignore some
of the essential developments that have largely taken
place outside the Netherlands. In Section 3.1, we start
with a brief overview of the pharmaceutical industry
before the biotechnology revolution arrived. Next we
analyse the impact made by biotechnology along the
cycle of discovery, first the move from exploitation
to differentiation (Section 3.2), and then the move to
reciprocation and exploration (Section 3.3). Finally, we
discuss the implications of these sectoral dynamics for
the emergence of new organizational forms (Section
3.4). In this analysis, we do not present new empirical
evidence. Instead, we make use of some highly insight-
ful, secondary sources that have extensively described
the revolutionary changes in this industry, and to which
the reader is referred.
3.1. Before the biotechnological revolution
The core knowledge base of the pharmaceutical
industry until the early 1960s was formed by organic
chemistry. It had developed in Germany and the UK
in the late 19th century and had consolidated in differ-
ent types of dominant designs formed by a variety of
chemical production processes. These processes gener-
ated different types of products that, in generalisation,
were used for a wide range of applications ranging from
polymers, paints and coatings, lubricants, cosmetics,
and detergents to pharmaceuticals. It was this grow-
ing body of knowledge on and experience with organic
chemistry that increasingly enabled to produce phar-
maceuticals on a large-scale and in a reliable way. As far
as such pharmaceutical applications were concerned,
organic chemistry provided the possibility to develop
an in-depth understanding of the chemical properties
of molecules and how they interact with one another.
However, organic chemistry did not provide an under-
standing of the human body. As a consequence, the
prevailing industry recipes and mindsets were focused
on understanding the properties of chemical entities
and producing them reliably in large quantities (Santos,
2003), not on understanding the biological underpin-
nings of specific diseases. At best, organic chemists
could develop a sense of a relation between a chemical
compound and its potential therapeutic effect (Pisano,
2002). These chemists therefore tended to focus on the
synthesis of chemical compounds that had already been
shown to have positive effects. These compounds were
7
then screened for their therapeutic potential. So, the
discovery of new drugs was a mainly tacit capability
that was highly dependent on the skills of individual
chemists, which made it difficult to codify. Still, this
search process of random screening worked very well
for a large number of years and generated important
classes of drugs, making this period after WWII the
‘golden age’ of the pharmaceutical industry (McKelvey
and Orsenigo, 2004; Pisano, 2002). In terms of our
cycle of discovery, we can characterise this period as
one with a strong focus on incremental innovations
and the exploitation of the existing knowledge base
of organic chemistry.
3.2. From exploitation to differentiation
The advent of biotechnology had an impact on
this knowledge base of organic chemistry in so far
as its application in pharmaceuticals was concerned.
In the early 1980s, some first signs emerged that
things were changing. The massive public funding for
health-related research after WWII began to bear fruit.
Substantial progress in scientific areas such as phys-
iology, pharmacology, enzymology and cell biology
created a growing understanding of the biochemical
and molecular roots of diseases and of the effectiveness
of existing drugs in curing these diseases (Jungmittag
et al., 2000). These new medical insights in diseases
offered researchers new areas to apply and develop
their skills and to diversify the development of new
drugs. The first biotechnological revolution created a
branching of the knowledge base of organic chemistry,
which was formed by a combination of organic
chemistry and molecular biology. So, the established
practice based on organic chemistry remained, whereas
molecular biology provided novel elements into this
practice. Although the core knowledge base remained
firmly rooted in organic chemistry, this new biological
knowledge enabled chemists to take a more rational
approach to the design of new drugs: it enabled them
to define the search space more accurately and to take
a more structured approach to the screening process.
So, random screening turned into a more guided
search process (McKelvey, 1997; Pisano, 2002).
During this period, those pharmaceutical firms that
had maintained absorptive capacity through in-house
R&D and developed close relations with networks of
individual scientists were able to make this transition
8 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
(Gambardella, 1995; Pisano, 2002; Santos, 2003;
McKelvey and Orsenigo, 2004). As a consequence,
the productivity of the existing search process of these
firms improved substantially (Nightingale, 2000). The
advent of molecular biology enabled pharmaceutical
firms to now fully exploit the possibilities of the exist-
ing knowledge base of organic chemistry (McKelvey
and Orsenigo, 2004). In terms of our cycle of discovery,
these empirical findings fit with the phase of differenti-
ation as it was the established practice of search based
on organic chemistry to which novel elements (molecu-
lar biology) were added. These novel elements yielded
incremental adaptations to this established practice,
namely a more rational approach to search and screen-
ing. So, in differentiation we see that novel elements
enabled a more scientific understanding of diseases
and that the ‘trial-and-error’ nature of the search
process was increasingly being complemented by a
more structured approach. Still, basic design principles
were maintained as the random character of the search
process did not disappear and the sequential nature of
the search process of drug discovery remained in tact.
These extended efforts of exploitation of the existing
knowledge base of organic chemistry made its lim-
itations increasingly visible, first slowly but towards
the end of the millennium increasingly more apparent.
These limitations became manifest by the increasing
number of existing patents by pharma companies that
were expiring, in combination with a decreasing num-
ber of potential ‘blockbusters’ in the pipe-line.4,5 In
line with our theoretical analysis, this growing accu-
mulation of failures was indicative of the fact that
adaptations made in the phase of differentiation turned
out to be insufficient and did not further improve per-
formance. Based on our logic of the cycle of discovery,
when differentiation is not sufficient for adaptation, we
anticipate experiments with novel elements, adopted
from a novel context, which seem to be successful
where established practice fails. This is the phase of
4 An example is formed by Eli Lilly’s patents for its very successful
Prozac (against depressions) which expired from the beginning of
2001. The turn-over of US$ 2.3 billion in 2001 has halved in a few
months and could not be compensated for by increasing sales of other
new medicines (FD, January 2002).
5 According to Jan Leschly, former CEO of SmithKline Beecham,
“Big pharma cannot afford to rest on its laurels. If today’s successful
pharma companies do nothing then their sales will be halved within
10 years from now” (Ernst and Young, 2001a).
reciprocation, consisting of a hybrid structure of exist-
ing and new elements that enables to test the potential
of these novel elements, and in which basic design prin-
ciples are increasingly being questioned.
3.3. Reciprocation and beyond
In the late 1980s and early 1990s, a second wave in
the molecular biological revolution emerged: genetic
engineering. This second revolution was more radical
than the first revolution as it opened up completely new
areas for innovation and altered the drug discovery pro-
cess in profound ways. Genetic engineering forms an
umbrella-name for a newly emerging set of complex
and multidisciplinary technologies, formed by among
others recombinant DNA (rDNA), monoclonal anti-
body technology, gene therapy, and later combinatorial
chemistry and high throughput screening. The appli-
cation of these new knowledge bases followed two
different types of technological trajectories. One was
the application of these genetic engineering techniques
as a process tool to produce proteins, the other trajec-
tory consisted of a combination of genetic engineering
and molecular biology and acted as a research tool to
enhance the speed and efficiency of the discovery pro-
cess of new drugs.
The first trajectory opened up a completely new
perspective as its focus was on new process devel-
opment that enabled to produce proteins synthetically
in large(r) amounts that been unavailable for com-
mercial use until then (Henderson et al., 1999). Most
of these proteins were based on natural or modified
human proteins and their therapeutic qualities were
well understood through organic chemistry (Henderson
et al., 1999; McKelvey and Orsenigo, 2004). In terms
of the logic of the cycle of discovery, this reflects a
move to the phase of reciprocation as a hybrid prac-
tice was created that was built up of existing elements
(understanding still based on organic chemistry) and
new elements (genetic engineering), and that formed
a topple point between exploitation and exploration:
it was exploitation of the existing knowledge on the
curing effects of familiar proteins, it was exploration
of new processes that enabled to produce (existing
and new) proteins. However, it also revealed a con-
straint on realising this new potential, namely the lack
of a fundamental understanding of the engineering of
biotechnology-based processes. Not only was such a
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
basic scientific understanding lacking, the longstand-
ing experience in the industry of developing chemi-
cal processes but also appeared irrelevant as far as
the development of such biotechnology-based pro-
cesses was concerned (Henderson et al., 1999). As
a result, established large pharma firms did not have
the required capabilities to further explore this route,
which was therefore mainly explored by new entrants
(McKelvey and Orsenigo, 2004). To reduce complex-
ity and risks, most new entrants focused therefore on
producing existing proteins that were already in use
as drugs or proved to be difficult to upscale through
traditional chemical processes. However, they ignored
the development of new ones (Henderson et al., 1999).
As a consequence, scepticism emerged whether rDNA-
based processes could be scaled-up at all. In the early
1990s, therefore, it became clear that the avenue taken
by this first trajectory was more expensive and difficult
than anticipated. Moreover, the new entrants pursu-
ing this trajectory lacked competencies in the field
of clinical testing and regulatory approval procedures.
As a consequence, this trajectory exhausted over time,
increasingly yielding diminishing returns (Henderson
et al., 1999; McKelvey and Orsenigo, 2004).
The second trajectory had its focus on the search
and discovery of new drugs. The combination of
molecular biology and genetic engineering techniques
opened up a fundamentally new way of discovery:
it enabled to identify upfront clearly defined search
spaces, potential targets and effective heuristics. This
new search method that emerged was based on a ‘lock
and key’ philosophy of finding a molecule that binds
to a protein and turns it off, preventing the protein
from initiating disease causing effects (Nightingale,
2000). The challenge then was to find one or more
lead compounds that could function as such a ‘key’
that fits the protein as the ‘lock’. Whereas in the period
of random screening drug compounds were basically
‘discovered’, this new rational approach was aimed
at the ‘design’ of drugs, following the more scientific
understanding of the biological underpinnings of
diseases (Pisano, 2002). So, the random nature of
the search process, considered a limitation in the
phase of differentiation, was addressed explicitly by
introducing an approach of maximising rationality in
the search process through the incorporation of novel
technologies. In fact, the ultimate goal now became
to design new drugs as rationally as possible. In other
9
words, a basic design principle of exploitation, random
screening, was being sacrificed. As we argued in our
theoretical analysis, in the phase of reciprocation basic
design principles are increasingly being questioned,
in hybrid structures. Randomness, as a basic design
principle of the search process, was now being elim-
inated through maximising rationality as a new design
principle. In terms of the cycle of discovery, this was in
one step from differentiation to beyond reciprocation.
However, this proved to be too large a step.
In the early 1990s, it became increasingly clear that
this fully rational approach did not deliver satisfactory
results as it showed disappointing performance,
incurred high costs, and provided unattractive revenue
potential (Nightingale, 2000). The reasons were as
follows. To improve rationality as much as much
as possible, firms started to rely more and more on
substantial computing power that supported the search
process by producing accurate approximations of
chemical compounds and by enabling refined simula-
tions of their potential effects on diseases. The initial
expectation was that such simulations could replace
labour-intensive experiments, however, they appeared
to lack sufficient accuracy and also proved to be more
costly than anticipated (Henderson et al., 1999). More-
over, this rational and large-scale approach to drug
discovery also led to the development of increasingly
similar drugs across pharmaceutical firms, resulting
into little product performance differences and con-
sequently lower market potential (Nightingale, 2000).
Because of these reasons, maximising rationality in
the discovery process was increasingly considered as
unattractive. As a response, firms moved away from
this new approach and started to introduce again some
randomness into their search process in order to be able
to differentiate from competitors (Nightingale, 2000).
So, this newly established practice of maximising ratio-
nality of the screening process, forming a step beyond
reciprocation, was increasingly abandoned, whereas
firms started to search for possibilities to introduce
randomness back again. In fact, this was now recip-
rocation as an important element of the old practice
(randomness) was incorporated again into the search
process. In line with our logic of discovery, this resulted
into a hybrid practice of old elements (randomness)
and new elements (large-scale, parallel screening pro-
cess), whereas other elements of the old practice, such
as the knowledge base on organic chemistry and serial
10 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
testing procedures, were increasingly considered as
obsolete.
The key technology that enabled this shift away
from purely rational drug discovery was formed,
initially, by high throughput screening technologies
(HTS), which enabled to combine a rational and a ran-
dom approach (Henderson et al., 1999; Nightingale,
2000; McKelvey and Orsenigo, 2004). Whereas the
original process of random screening was fairly labour
intensive, this technology of HTS was an automated,
robotics-based process that enabled the in vitro (in
stead of in vivo) testing of large numbers of com-
pounds. It represented a transition from the sequential,
craft-based search process to a parallel, large-scale
screening process by automated robots.6 In essence,
HTS is a random screening technology like tradi-
tional labour-intensive screening, but much faster and
cheaper (Pisano, 2002).
Following our logic, increasing experience with
such a hybrid structure in reciprocation yielded insights
in both its potential and in the constraints imposed
on the realization of that potential. Insights into the
potential indicated the possibility of a more funda-
mental understanding of diseases and their underlying
mechanisms, and the possibility to attack these causes
instead of curing symptoms of those diseases, as was
customary practice (Nightingale, 2000; Pisano, 2002).
Constraints were formed by limited possibilities for
synthesizing new compounds rapidly and efficiently
and for handling the vast amounts of data generated by
HTS. The data requirements could be handled by the
increasing availability of computing power. The other
constraint, possibilities to rapidly synthesize large
amounts of compounds, required a new synthesis pro-
cess. It was here that the first trajectory with its focus on
process development came into sight again. Whereas
in the second trajectory the focus was on drug devel-
opment and building a fundamental understanding of
their potential effects, testing them at a large-scale
required process development capabilities that had
been explored in the first trajectory, mostly by new
entrants. So, throughout the 1990s the two, initially
6 In the 1980s compounds were sequentially checked for toxicity,
absorption, carcinogenicity and so on, often consuming numbers of
years. Towards the second-half of the 1990s, this testing procedure
had turned into a parallel process in which all checks were determined
simultaneously (Sykes, 1997).
separated, trajectories have converged, leading to the
development of combinatorial chemistry (Henderson
et al., 1999; McKelvey and Orsenigo, 2004). This tech-
nology enabled pharmaceutical firms to produce large
numbers of different compounds in a reliable and cost-
effective manner (Gordon et al., 1994; Pisano, 2002).
So, HTS in combination with combinatorial chemistry
enabled a return to random screening, although in a
way that was much quicker and more productive than
before (Nightingale, 2000; Pisano, 2002; McKelvey
and Orsenigo, 2004).7 This shift from a sequential
process of screening hundreds of compounds towards
a parallel process of screening ten-thousands of
compounds, generated vast amounts of data. A proper
interpretation of these data then required state-of-the-
art software in the field of database, datamining, data
management, statistical analysis, visualisation tech-
niques, and so on (Gambardella, 1995; Nightingale,
2000).8 As a consequence, the clear potential of
combining both trajectories provided an incentive and
insight for a more architectural radical change beyond
reciprocation, in the phase with a focus on exploration
of a novel structure of old elements (randomness)
and new elements (recombinant DNA, monoclonal
antibody technology, HTS, combinatorial chemistry,
and state-of-the-art software technologies).
3.4. From technology to organization: networks as
new organizational forms
Following our logic along the cycle of discovery, the
move beyond reciprocation makes that existing organi-
zational forms no longer suffice and need restructuring
in order to open up to new variety of content. In line with
these theoretical claims, the empirical evidence clearly
indicates that to realise the new potential invoked by
the radical change beyond reciprocation, new organi-
zational forms were needed.
Without exception, new entrants formed by small
start-ups that specialised in biotechnological research,
have been at the leading edge in transforming the
7 This random character is still present, even with new drugs that
are introduced at today’s markets. Pfizer’s Viagra was initially devel-
oped as a treatment for hypertension, which proved to be limitedly
successful. However, it has become an enormously successful drug
for erectile dysfunctionings.
8 Analysis of these data through such databases is also referred to
a as ‘in silico analysis of populations’ (Nightingale, 2000).
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
highly scientific knowledge created at universities into
potentially commercially useful techniques and prod-
ucts (Reiss et al., 2000; McKelvey and Orsenigo, 2004;
Pisano, 2002; Roijakkers, 2003).9 At the same time
though, these new entrants have not wiped out incum-
bent firms, on the contrary. The large cognitive dis-
tance between the newly developing knowledge base
on biotechnology and the existing knowledge base of
organic chemistry (Powell et al., 1996), as well as
the profound differences between old and new search
routines, has made the successful adoption of the
new technologies difficult for incumbents, although
this has varied substantially among them (Reis et
al., 2000; Roijakkers, 2003; Santos, 2003; McKelvey
and Orsenigo, 2004). Especially, firms that adopted
the first molecular techniques in the early 1980s and
that made the subsequent transition from random to
guided drug discovery proved to be successful in man-
aging this change process (Zucker and Darby, 1997;
Henderson et al., 1999; Pisano, 2002; McKelvey and
Orsenigo, 2004). These firms had developed strong
in-house R&D-capabilities and nurtured close links
to scientists that enabled them to build up and main-
tain sufficient absorptive capacity and identify relevant
external knowledge (Gambardella, 1995; Henderson et
al., 1999; Reis et al., 2000; Santos, 2003).10 These
continuous investments in such external links and in
in-house R&D substantially increased the costs of the
search process, whereas firms initially believed that the
new techniques would reduce these costs (Nightingale,
2000). Although a costly affair, firms that had not made
this initial transition nor had developed relations with
scientists, were often very slow followers and often
proved to be unsuccessful in adopting the new tech-
niques (Gambardella, 1995; Pisano, 2002).
So, the biotechnology revolution has proven to be
competence enhancing for those large pharma firms
that timely changed their search routines and adapted
9 In the literature, these small specialised firms are often referred
to as Dedicated Biotechnology Firms (DBFs). These DBFs connect
a ‘basic scientific environment’ with its emphasis on the importance
of new knowledge with a ‘techno-economic environment’, which
emphasizes economic value (McKelvey, 1997). As such they form
a new type of boundary organizations that can be labelled as ‘firm
laboratories’ (Callon, 2002).
10 In a study on this topic, Gambardella (1992) found that in-house
scientific capabilities importantly explained firm differences in their
R&D productivity, when controlling for firm size.
11
their organization through close links to science and
small specialised firms (Pisano, 2002; Santos, 2003).
As a consequence, the new industry recipe that emerged
throughout the 1990s and basically still prevails until
today, was that new entrants specialised in the new and
more radical technologies such as genomics, gene ther-
apy, combinatorial chemistry and son on. In doing so,
(most of) the new entrants did not aim to become drug
producers themselves, but rather acted as providers of
these new technologies that formed the research tools
in the drug discovery process of the large pharma firms.
As a consequence, the capability of these large pharma
firms to access and absorb such external knowledge has
increasingly become a key success factor in the indus-
try (McKelvey and Orsenigo, 2004). In general, drugs
that were developed in-house had a much higher prob-
ability of failure than drugs of which development was
based on the use of external licences (Gambardella et
al., 2000). So, the new entrants have functioned as the
‘transfer channel’ of new technology from science to
established pharma firms. At the same time, these large
pharma firms disposed over the capabilities to deal with
(clinical) testing and regulatory approval procedures,
which the new entrants basically lacked.
In sum, these empirical findings, as extensively
described in the literature, are in line with our theo-
retical claim: in order to be able to realise the potential
of the novel combination, the existing organizational
form (in-house R&D) no longer sufficed and needed
restructuring to open up to cognitive variety (scientific
knowledge accessed and recombined by new entrants).
As analysed, this network has proven to be an effective
organizational form for dealing with the new chal-
lenges, making it a more useful concept for analyzing
innovation than firms (Powell et al., 1996). Therefore,
we further analyze how these sectoral dynamics have
settled in different types of network structures. In doing
so, we will focus on the Netherlands.
4. Understanding networks in the Netherlands
In this section, we will focus on developing an
understanding of the networks that have emerged in
the Netherlands in the late 1980s towards the early
years of the new millennium. Although the literature
abounds with claims and empirical evidence on the
role of networks in this industry (Gambardella, 1995;
12 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Powell et al., 1996; Henderson et al., 1999; Orsenigo
et al., 2001; McKelvey and Orsenigo, 2004), a more
detailed understanding of how they are structured and
of their inner-functioning is still underdeveloped. The
aim here is therefore to open up this ‘black box’ by ana-
lyzing these networks along two lines: (1) density and
strength of the interfirm relations that make up these
networks, and (2) a combined competence and gov-
ernance perspective. To analyze networks in this way
has not been done yet in the literature, which makes
it not a straightforward task. Therefore, to be able to
analyze networks along these lines and not to make our
analysis unnecessarily complex, we abstract from spe-
cific issues that may be typical for Dutch firms in this
industry. This choice makes our analysis potentially
also more relevant for understanding similar networks
in other countries.11
In the Netherlands, the majority of the ‘new
breed’ of Dedicated Biotechnology Firms (DBFs) were
engaged, through contract research services, in gen-
eral platform technologies (see also Appendix A). In
doing so (most of) the new entrants did not aim to
become drug producers themselves, but rather acted
as providers of these new technologies that were used
in the drug discovery process of the large pharma firms.
Specialisation in platform technologies has the benefit
that in general time-to-market is shorter and that there
are less risks involved as compared with the “block-
buster” model. This latter often requires lengthy and
costly clinical trials with a higher chance of failure, but
also with potentially very high revenues if successful.12
11 Various empirical studies have indicated that there are some
major invariances across countries in the properties of innovation pat-
terns of the biotechnology sector (Marsili, 2001). The links between
science, DBFs and large pharmaceutical firms seem to be such a
general property of the biotechnology industry. Although a sectoral
characteristic as such, the strength of these links varies notably across
countries, with the strongest links in the US (Henderson et al., 1999;
Pisano, 2002). Such growing similarities seem to appear because
countries increasingly imitate ‘best practices’, as well as due to the
global nature of the sectoral changes in markets, demand and com-
petition in the pharmaceutical industry (McKelvey and Orsenigo,
2004).
12 In this respect, Dutch DBFs tend to resemble German biotech-
nology firms, most of which have also specialised in general platform
technologies. This is not so surprising given the fact that the national
instititutional framework of the Netherlands resembles the German
context, sometimes also referred as the ‘Rhine-land model’. In con-
trast, the UK, for example, tends to specialise in the therapeutics
Fig. 2. Emerging knowledge value chain and learning regimes in the
field of general platform technologies.
Given this, we will further focus on DBFs in these
general platform technologies. In this field, a ‘knowl-
edge value chain’ has emerged throughout the 1990s,
as schematically depicted in Fig. 2:
Within this value chain we can discern between two
main types of learning regimes, namely:
- Learning regime 1: Focus on exploration (embedded
within a network of DBFs with academia).
- Learning regime 2: Focus on exploitation (embedded
within a network of DBFs and large pharma).
Concerning sources of data, we needed information
on a broad range of issues, from data on industry
characteristics, on structural and relational embed-
dedness and on interfirm learning and innovation
activities. In this respect, our primary source of data
is formed by industry reports and studies. Given the
(growing) importance of the biotechnology industry
to the Dutch economy, it has been extensively studied
over the past 10–15 years. These studies were carried
out by individual (scientific) researchers, specialised
research institutes and consultancy firms. Clients
of these studies ranged from the Dutch Ministry of
Economic Affairs,13 the Dutch Ministry of Transport
and Telecommunication, various industry associations
(e.g. Biopartner, Nefarma), the OECD and EU. Most
of these reports cover (some of) the issues that we
are interested in and, taken together, these multiple
sources enabled us to develop an approximation
of our network constructs over the late 1980s and
throughout the 1990s. Using a variety of industry
segment, with high risks and possibly high gains, which reflects its
more liberal and market-based institutional framework (Casper and
Kettler, 2001).
13 At the time, the first author was employed at this Ministry and
acted as the principal coordinator for such industry studies. His role
was to monitor these studies in terms of progress and to maintain
quality standards. Moreover, such progress reports were regularly
discussed with firms in the field and/or with industry watchers at
meetings during which the first author was also present.
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
reports also enabled us to triangulate among them. In
addition, we have approached two recognized industry
experts with our analysis. We asked them to check
whether our analyses were correct in terms of facts
and completeness and whether key information was
missing. Most of the innovation dynamics in this
industry have taken place in a relatively short period
of time, with, as far as the Netherlands is concerned, a
major chunk throughout the 1990s and the early years
of the new millennium (EZ, 2000; Enzing et al., 2005).
This enabled us to study (relatively) recent events,
enhancing the reliability of our experts’ judgement. As
a result, this combination of data collection methods
(industry reports and experts’ judgement) has enabled
us to also triangulate between these two sources, which
is important for internal validity and reliability.14
There are two methods of identifying networks
(Knoke and Kuklinski, 1982). A realist approach is
based on the subjective perception of the involved
actors. The identification of a network is determined
by ‘the limits that are consciously experienced by all
or most of the actors that are members of the entity’
(Knoke and Kuklinksi, 1982, p. 22). A nominalist
approach is based on the viewpoint of the researcher.
Identification of a network is based on the application
of his analytical framework used. In this respect,
we have followed a nominalist approach by only
considering those networks with relevance from an
innovation perspective.
We now further analyse each learning regime more
in depth: in Section 4.1, we discuss a network with a
focus on exploration (learning regime 1) and in Section
4.2, we discuss a network with a focus on exploitation
(learning regime 2). Furthermore, in Appendix A, we
provide more background information of the develop-
ment of the pharmaceutical biotechnology industry in
the Netherlands from the late 1980s onwards.
4.1. Learning regime 1: Technological exploration
Learning regime 1 was embedded in a network made
up of relations between DBFs and (public) research
14 For the analysis of networks we have studied 26 reports and
22 articles. We have approached Christien Enzing of TNO and
Sander Kern of Dialogic who have both conducted numerous stud-
ies on biotechnology for the Dutch ministry of Economic Affairs,
the EU and the OECD (Paris). See for more information on Enzing
www.stb.tno.nl and for more information on Kern www.dialogic.nl.
13
institutes, and its main focus was on exploring new
knowledge. We distinguish between two periods, one
period from the late 1980s towards the late 1990s and
a second period from the late 1990s onwards.
4.1.1. From the late 1980s towards the late 1990s
From the 1980s towards the middle of the 1990s,
the knowledge base on general purpose platform
technologies had a mainly stand-alone nature due to
its strong basis in molecular biology and/or genetic
engineering technologies. In this period, there were
no DBFs created through divestment from existing
pharmaceutical firms, which was indicative of the
large cognitive distance with the existing knowledge
base of organic chemistry. Also, the unwillingness
of the existing labour force of organic chemists to
‘give up’ their powerful positions to pharmacologists
and biotechnologists prevented incumbent firms from
jumping onto these new opportunities (Degenaar and
Janszen, 1996). Following the highly science-based
nature of knowledge, the majority of DBFs cooper-
ated intensively with (public) research institutes (EZ,
1998). The learning outcome of this search process was
formed by abstract and codified knowledge. The search
process of scientific discovery itself was characterised
by a lot of trial-and-error and was highly specific to
individual persons and research communities (Enzing,
2000). This process entailed many elements that were
difficult to codify, such as test set-up, accurate exe-
cution, interpretation of test results, and so on. It was
characterised by serial, incremental improvements,
leading to the accumulation of tacit knowledge within
stable research groups of academics and DBFs (Enzing
and Kern, 2002). So, a science-based and fast-changing
knowledge base developed, with cumulative charac-
teristics and high specificity to a network of DBFs
and academics (Enzing and Kern, 2002). Relations
between these people were dense and of fairly high
durability (4–5 years or more) with frequent inter-
action (varying between daily and weekly contacts,
van Geenhuizen, 1999) in mutual openness on mainly
search-related and technological issues. Coordination
took place through a combination of contracts and
social mechanisms. Contracts were used to arrange
economic aspects such as fees for using scientific staff
(fte’s) and laboratory facilities as well as allocating
intellectual property rights to the people involved.
Next to this, coordination was mainly social in terms
14 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Table 2
Learning regime 1—Technological exploration
Late 1980s to late 1990s
Properties of knowledge base (competence)
Highly tacit (search process)
Highly codified (search output)
Low systemicnessa
Medium rate of change
Governance
Contracts, mutual self-interest, peer reviews, reputational control
Networks
High density, small size, informal, inflexible ties, locally embedded
Strength of ties
High frequency of interaction, long duration, high openness, low
cognitive distance
a Systemic knowledge refers to the extent in which knowledge is built up of an integration of (1) different scientific and/or engineering
disciplines, or (2) different competencies such as R&D, manufacturing, marketing, and so on (Teece, 1986; Malerba and Breschi, 1997).
of peer reviews and reputational control in view of
academic quality standards (van Geenhuizen, 1999).
In this network, a clear spatial concentration could
be observed, especially around universities in Amster-
dam, Groningen, Leiden, Utrecht, Nijmegen, Wagenin-
gen, Maastricht, and Delft. The mainly tacit search
process meant that personal contacts and frequent inter-
action were necessary to accommodate an effective
transfer of this tacit knowledge (van Geenhuizen and
van der Knaap, 1997). In addition, physical proxim-
ity facilitated easy access to a talent pool of skilled
workers, facilitating knowledge spill-overs through the
mobility of researchers. In addition, opportunities were
generally diffuse, requiring regular checks, and adap-
tations of the search process into the most promising
search direction (van Geenhuizen, 1999).15 The impor-
tance of physical proximity was further indicated by
the fact that most patents are assigned to inventors from
within the Netherlands (Allansdottir et al., 2002) in this
15 These findings are in line with a recent study on biotech firms
in Sweden. In this study it was found that geographical proximity
was especially important firms–university cooperation, much less for
firm–firm cooperation or university–university cooperation (McK-
elvey et al., 2003).
Late 1990s-onwards
Highly tacit (search process)
Highly codified (search output)
Medium to high systemicnessa
Medium to high rate of change
Network core: contracts, mutual self-interest, peer reviews
and reputational control
Periphery: contracts (mostly licenses)
Network core: high density, small size, informal and flexible
ties, locally embedded
Periphery: low density, high entry and exit, delocalised
Network core: high frequency of interaction, long duration,
high openness, low cognitive distance
Periphery: low frequency of interaction, short-medium dura-
tion, limited openness
period. So, these exploration networks were formed by
dense networks of strong ties between universities and
DBFs that enabled them to develop an in-depth under-
standing and critical peer reviews (Enzing and Kern,
2002) (see also Table 2).
4.1.2. From the late 1990s and onwards
Towards the end of the 1990s, platform technolo-
gies increasingly became more multidisciplinary in
nature and entailed varying combinations of disci-
plines. Examples of such systemic platform technolo-
gies are DNA chips, rDNA arrays, proteomic analysis
combining electrophoresis, and NMR, and genomic
sequencing (Degenaar and Janszen, 1996; Enzing,
2000). In addition, combinations with software and
microelectronics started to develop that allowed for
miniaturisation, automation, and data-mining. This
was also reflected in a growth of the number of
patents with multiple assignees, especially over the
second-half of the 1990s (Allansdottir et al., 2001). In
this respect, opportunities increased, which was also
reflected by the fact that the number of entrants further
accelerated to 22 in 2000 (Biopartner, 2001), providing
further substance to the emerging Dutch pharmaceuti-
cal biotechnology industry.
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Due to the mainly monodisciplinary orientation
of their academic partners in this learning regime 1,
DBFs have also actively started to search and access
complementary (scientific) knowledge, wherever it
was located. As a result these dense networks of strong
ties between universities and DBFs were opening up to
complementary, outside sources of knowledge. Such
outside sources were formed by universities, research
institutes and other networks outside the Netherlands,
either at various locations in the US, the UK, and Ger-
many (Enzing et al., 2003). Because this knowledge at
universities and research institutes was codified, it was
easily accessible and transferable by means of publica-
tions or Internet (Ernst and Young, 2002). Especially
the use of Internet enabled DBFs to share information
with anyone around the globe and to access public
databases with state-of-the-art (scientific) knowledge.
These relations relied much less on geographical
proximity and could take place over (very) long
distances (van Geenhuizen and van der Knaap, 1997).
So, these more virtual linkages with these distant
created a non-dense periphery that was complementary
to these dense, local research networks. Relations in
this periphery were generally of low strength in terms
of duration, frequency and entailed a limited set of
specific technological issues (Enzing and Kern, 2002).
The division of partners over the category ‘local’ versus
‘distant’ ties was approximately 50–50%. Although
the periphery was made up for 50% of such distant ties,
the share of licensing agreements and formal R&D col-
laborations was small (approximately 10%) indicating
the high volatility of these ties (Enzing et al., 2005)
(see also again Table 2). Governance and coordination
of these relations took place in various forms such as
licenses, technology partnerships or by research con-
tracts with scientific organizations. This also explained
the large inflows of knowledge from US-based licences
into the Netherlands (Degenaar and Janszen, 1996;
Enzing et al., 2003) (see also Table 2 again).
4.2. Learning regime 2: Technological
exploitation
Learning regime 2 was concerned with exploitation,
more than with new knowledge creation (Degenaar and
Janszen, 1996; Ernst and Young, 1999). We study this
learning regime from the late 1980s towards the late
1990s towards the early years of the new millennium.
15
Learning regime 2 was embedded in a network made
up of relations between DBFs and large pharma firms
(see Fig. 2 again). The rationales underlying this net-
work were as follows. Large pharma firms needed to
keep up to date with a rapidly changing knowledge
base that was also diverse and systemic, built up from
various disciplines. At the same time, opportunities
pertained to niches and were also diffuse, making it
difficult for pharma firms to decide in which fields
of knowledge to invest and which to ignore (Ernst
and Young, 2001a,b). Therefore, large pharma-firms
made use of alliances with various small DBFs, which
enabled them to explore various opportunities at the
same time, without making substantial specific invest-
ments. The attractive but diffuse nature of opportunities
and the fact that knowledge was highly specific to
these DBFs meant that network structure was made
up of mainly bilateral relations between large pharma-
firms and DBFs. For these DBFs, a core performance-
yardstick is ‘time-to-patent’ (Degenaar and Janszen,
1996; Ernst and Young, 1999, 2001c.
This combination of attractive, diffuse opportunity
conditions and a rapidly changing knowledge base
required diversity in search spaces. As a result, a net-
work structure emerged that combined a variety of
decentralised approaches with the ability to coordi-
nate this diversity in a fairly light way. This net-
work structure also explains why in most cases DBFs
did not cooperate closely together when they were
already cooperating with a large pharma-firm. To inter-
act directly with one another would require specific
cognitive investments from the involved DBFs. DBFs
were highly specialised in a specific technological field
and given the stand-alone nature of knowledge, poten-
tial linkages among these specialised fields were absent
or very limited (Degenaar and Janszen, 1996). This did
not provide an attractive perspective for a sufficient
duration of such a relation in order to recoup such
investments. As a consequence, spill-overs between
DBFs were very limited (EZ, 1998; Senker, 1998) (see
also Table 3).
An important governance instrument in this
network was formed by research contracts, possibly
complemented by minority equity arrangements (Ernst
and Young, 1999; Enzing, 2000). These contracts
regulated contract research, contract manufacturing,
custom synthesis, development, sequencing, testing,
software design, and so on. They aimed to provide
16 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Table 3
Learning regime 2—Technological exploitation
Late 1980s towards 2001, 2002
Properties of knowledge base (competence)
Highly tacit (search process)
Highly codified (search output)
Medium systemicnessa
Medium to high rate of change
Governance
Contracts
Network position
Trust-in-competence
Networks
Low density
Small size
Open, structural holes
Locally disembedded
Strength of ties
Low frequency of interaction
Medium duration (2–5 years)
Low openness
High cognitive distance
a Systemic knowledge refers to the extent in which knowledge is
built up of an integration of (1) different scientific and/or engineering
disciplines, or (2) different competencies such as R&D, manufactur-
ing, marketing, and so on (Teece, 1986; Malerba and Breschi, 1997).
some duration to the relationship and to assure the
DBF of sufficient resources (especially money) to
do research. In general, these contracts were being
evaluated every 2–3 years (EZ, 1998).
The case of Dutch DBFs specialised in general plat-
form technologies now raises some interesting ques-
tions, such as: Why were contracts such an important
coordinating mechanism? How were these contracts
monitored given the high rate of change induced by the
exploration activities in learning regime 1? And how
did the cognitive distance between large pharma firms
and DBFs affect the need for specific cognitive invest-
ments?
In essence, the contract regulated a relatively
arms–length relationship between two parties. A DBF
devoted a lion’s share of its resources to explore a spe-
cific technological field to which the large pharma firm
contributed by financial resources. From the viewpoint
of a large pharma firm, it was basically subcontracting
research to a DBF, and its interest mainly concerned
the codified knowledge that came out of this search
process, not so much this process itself. This was an
important point: although the cognitive distance with
its existing knowledge base of organic chemistry was
fairly large, a pharma firm did not need to engage
in specific cognitive investments (such as mastering
all of these tacit competences and search routines)
nor in the build-up of trust.16 This also explained the
possibility of cooperating without the need of being
physically close. These factors then made contracts a
rational choice. Its value lay mainly in regulating a
‘first-right-of-refusal’ for the large pharma firm that
contributed the costs of doing research, manufacturing
and marketing. This entailed the rights to have first,
exclusive access to the results upon which he could
decide whether to use these or not. The fact that knowl-
edge as an outcome of the search process was highly
codified explained why contracts could be used in the
field of general platform technologies (and in biotech-
nology in general) as it allowed a direct assessment
of the amount of progress being made. The high rate
of change explained the importance of evaluation on
a regular basis. Depending on the outcome of such an
evaluation, the contractual relation was durable as long
as opportunities proved to be viable. If not, relations ter-
minated and parties separated. This easy break-up was
possible due to the fact that specific cognitive invest-
ments were limited and trust-in-intention was not a key
coordinating mechanism (Roijakkers et al., 2005).
5. Comparing empirical findings with
theoretical claims and conclusions
In this section, we discuss whether our logic of
the transition between exploitation and exploration at
a sectoral level, and the organizational implications
following from that, are supported by our empirical
analysis of the biotechnological revolution. To do so,
16 Although large pharma firms did not need to make such highly
specific, cognitive investments, they needed to constantly invest in
their absorptive capacity through in-house R&D. This enabled them
to develop and maintain a broader, more general perspective than the
specialised DBFs. Such a broader perspectice was required in order to
be able to develop new drugs that were marketable after all, as well as
to assess the value of new investment proposals by DBFs. Following
McKelvey and Orsenigo (2004), this broader orientation of large
pharma firms formed another reason why this division of labour with
DBFs persisted, apart from the complementary assets argument of
dealing with the complexities of product approval procedures.
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
we discuss our findings on learning regime 1 (Section
5.1) and learning regime 2 (Section 5.2). As a final step
we relate the empirical findings to the general logic of
our cycle of discovery and discuss in how far it has
served as an adequate theory of understanding the tran-
sition between exploitation and exploration. Based on
that we conclude (Section 5.3).
5.1. Learning regime 1: Technological exploration
Learning regime 1, as it developed in the first
period from the mid 1980s towards the mid-late 1990s,
focused specifically on the development of an in-depth
understanding of stand-alone, science-based knowl-
edge. This search process was embedded in relations
of fairly high durability with frequent interaction in
mutual openness on mainly search-related and techno-
logical issues. In order to deal with this complexity,
a dense network of strong ties emerged, coordinated
by a combination of research contracts, mutual self-
interest and social mechanisms such as peer control
and review. As argued, the research contract arranged
some key economic aspects and the allocation of intel-
lectual property rights, but beyond that, coordination
was mainly social in terms of peer reviews and reputa-
tional control in view of academic quality standards.
These findings on learning regime 1 do provide sup-
port for our claim that exploration requires a dense
network. In contrast to our earlier hypothesis on tie
strength is the finding that relations were strong in dura-
tion and partly coordinated by contracts, as a comple-
ment to social coordination. The role of contracts can
be explained by the fact that they specified the inputs
to the search process (in terms of fte’s and the use of
laboratory facilities) and ownership issues with regard
to the outcomes of the search process. Such capacity-
related inputs can indeed be specified upfront, whereas
the codified nature of knowledge as outcome of the
search process enabled to specify allocation of intellec-
tual property rights. As argued, the search process itself
was highly tacit and was therefore coordinated through
social mechanisms. According to our arguments, as set
out before, such durable and to some extent formally
coordinated ties would yield insufficient variety for
exploration. Interestingly, these localised, dense net-
works were complemented, from approximately the
late 1990s onwards, by a non-dense periphery of ties,
outside the Netherlands, with generally low strength
17
in terms of duration and stability. These distant ties
created the required diversity at the input-side of this
learning regime, i.e. diversity in sources of knowl-
edge. As analysed, these distant ties were coordinated
through licences that enabled to access and use this
distant, codified knowledge without the need for sub-
stantial specific cognitive investments and the build-up
of trust.
This is in contrast with our prediction that in
such exploration-oriented networks coordination takes
place through informal mechanisms such as reputation,
trust-in-competence, and mutual self-interest. The
explanation for this is that knowledge held by these
distant ties, at often distant locations, is highly codified
through publications and patents. Its potential value
can therefore readily be assessed given the generally
high absorptive capacity present in the dense, local
network. The high rate of change of knowledge made
that such distant sources of knowledge succeeded
one another on a regular basis, which required the
constant monitoring for new potential sources. This
relatively high turn-over of such distant ties was
possible because the knowledge base was of a mainly
stand-alone nature. Substantial technological interde-
pendencies were absent or weak, so that these distant
ties could be replaced without the risk of creating
bottlenecks in adjacent technological areas. So, the
locus of diversity of knowledge was in the periphery
of the network, whereas selection and triangulation
took place in the core of the network, with more dense
and stronger ties. Hence, the periphery compensated
for potential dynamic inefficiencies of the core.17
5.2. Learning regime 2: Technological
exploitation
This learning regime formed the transition from
exploration to exploitation. As predicted for exploita-
tion networks, we find a non-dense network, formed
by a hub-and-spoke structure with the pharma firm
17 This is in line with some recent findings by Aharonson et al.
(2004) on the pharmaceutical biotechnology industry in Canada
as well as with findings by Owen-Smith and Powell (2004) on
the biotechnology cluster around Boston. Both studies indicate that
biotech firms operate within a geographically embedded cluster from
where they maintain large numbers of strategic alliances, many of
them at long distances.
18 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
in the core of the network. The stand-alone nature of
knowledge and the strong specialisation of DBFs made
that there were virtually no relations among them. This
network structure enabled a large pharma firm to gain
access to scientific research. These bilateral relations
between the central pharma firm and DBFs created
a non-dense network structure made up of relations
that were of limited strength and mainly coordinated
through contracts. As argued, contracts could be used
given the codified outcome of the search process.
These findings provide support for our prediction
that in exploitation non-dense networks are selected,
made up of relations that are strong in duration but show
less strength in terms of frequency of contact mutual
openness and breadth of issues covered. In addition,
support is also found for our claim that in exploitation
formal coordination mechanisms are selected such as
contracts and formal control mechanisms.
5.3. Conclusions
The focus of this paper is on a basic ‘system logic’
of not only how exploration builds on, but also shifts
existing systems of exploitation. Based on a cycle of
discovery, developed in earlier research, we discussed
this logic at a sectoral level and analysed how, in a new
round of radical innovation, sector-specific institutions
such as dominant designs, industry recipes, and mind-
sets changed. Furthermore, we analysed the effects of
such sectoral dynamics on organizational forms, in par-
ticular networks of firms. The question now is in how
far our cycle of discovery has served as an adequate
theory for explaining the sectoral transitions between
exploitation and exploration and the implications for
innovation networks?
In Section 3, we discussed the changes that have
affected the worldwide pharmaceutical industry. Fol-
lowing this analysis we can argue that these radical
changes can be well understood in terms of our cycle
of discovery. The different phases from exploration to
exploitation (consolidation, generalisation) and from
exploitation to exploration (differentiation, reciproca-
tion) could all be identified. Moreover, various key
claims made per phase could be substantiated such
as: (1) novel elements (molecular biology) added to
the established search practice of random screening
(based on organic chemistry) indicated the move from
exploitation to the phase of differentiation, (2) the accu-
mulation of failures within differentiation (indicated by
the lowering number of patents in the pipeline) that her-
alded the move to reciprocation, (3) the hybridisation of
the knowledge base in old and novel elements in recip-
rocation (importing genetic engineering technologies
while keeping randomness), and (4) the emergence of
new organizational structures (networks of scientists,
specialised new entrants, and large pharma firms) that
enabled to fully realise the new potential of novel com-
binations. Also in line with the analysis along the cycle
of discovery was the finding that it proved to be very
difficult to skip one or more phases. This applied to
two different levels. At a sectoral level, the approach of
maximising rationality in the search process reflected
a transition of moving in one large step from differ-
entiation to beyond reciprocation, which ultimately
failed. At a firm level, the empirical evidence indi-
cated that firms making the transition to the phase
of differentiation first and then reciprocation had an
advantage over firms who tried to move directly from
exploitation to full exploration, by skipping differen-
tiation and/or reciprocation. Another claim that could
be substantiated entailed the need to change existing
industry recipes on organization, i.e. to open up the
organization to new cognitive variety. As the empirical
evidence indicated, firms who had invested in external
links with new entrants and scientists and that devel-
oped drugs based on external licences, were far more
successful than firms that stuck to in-house develop-
ment (Gambardella, 1995; Gambardella et al., 2000).
From this, we may conclude that our logic has served
its purpose, although there are also new insights and
lessons for improving it. One new insight connects
with our claim that the experimentation with novel
elements in reciprocation yields insights in the con-
straints in realising the potential. Such insights may
invoke new innovations in adjacent areas that enable to
‘lift’ the constraints. However, this may not always be
practically possible such as the case of rDNA-based
technology, as a process-tool for the manufacturing
of proteins, indicated: the unavailability of scientific
and/or experience-based knowledge inhibited the up
scaling of this trajectory. In other words, insights in
constraints and motivation to lift them as such are
not enough and the ultimate consequence may be that
the realisation of the novel potential simply becomes
blocked. This trajectory of biotechnology as a process
tool yielded a second new insight, namely that also
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
in early exploration (in reciprocation) process innova-
tions can emerge, whereas we claimed that, in line with
the received wisdom in the innovation literature, pro-
cess innovations are generally to be expected towards
exploitation (in consolidation to enable the move to
generalisation). Another new insight is that we have
seen phases going on at the same time, indicating that
shifts between phases are not such a linear process as
described, but in fact reveal the possibility of over-
lapping phases or even the possibility of ‘jumps’: in
one step from differentiation to beyond reciprocation
as indicated by the approach of maximising rational-
ity in the search process. Still, our cycle of discovery
enabled us to decompose the complexity of these tran-
sitions and indicated that there is indeed a logic: one
step from differentiation to beyond reciprocation ulti-
mately proved to be a step too far and required a return
to reciprocation before being able to move on again.
There are also points for improvement. In our analy-
sis along the cycle of discovery, we have taken a rather
voluntaristic view on the motivation of firms to adapt to
change. As indicated, the willingness to change is not
spread equally over firms. A lack of willingness may
ultimately result in a lack of ability to change, with
major consequences for firms themselves. A lesson is
therefore that such a voluntaristic view holds for some
firms but certainly not for all of them. This connects
with a second lesson, which is that the case of pharma-
ceutical biotechnology provides further evidence for
the notion that change can indeed be imposed from
outside, instead of being voluntarily sought by incum-
bents. The described shifts in knowledge bases were
largely exogenous to the pharmaceutical industry in the
sense that molecular biology and genetic engineering
originated from search activities outside the involved
firms. These shifts have been endogenous in the sense
that their adaptation has largely occurred within (suc-
cessful) firms themselves and by establishing links with
new entrants.
A final point is as follows: whereas recent literature
stresses the idiosyncratic nature of the biotechnolog-
ical revolution, our analysis seems to reveal that this
may perhaps not be as unique as suggested. Based
on the logic of our cycle of discovery, it was pos-
sible to reconstruct and understand some of the sys-
tem logic that characterized the transitional process
between exploitation and exploration in this industry. In
this respect, our logic may contribute to an evolutionary
19
theory at a sectoral level and seems to have the potential
of generalisation across technologies and/or sectors.18
Such a theory has only started to emerge (Nelson, 1994;
Malerba, 2004).
Appendix A. Pharmaceutical biotechnology in
the Netherlands
In this appendix, we provide a brief overview of
the development of the pharmaceutical biotechnol-
ogy industry in the Netherlands from the late 1980s
onwards. In this way, this appendix serves as a con-
cise background to the analysis of the exploration and
exploitation networks as analysed in Section 4. Apart
from the performance of Dutch firms, both for explo-
ration and exploitation, we also provide some descrip-
tive statistics on Dutch DBFs.
A.1. Performance: exploration and exploitation
The track record of the Netherlands in the field of
pharmaceutical biotechnology is mixed. The position
seems particularly strong in exploration, i.e. scientific
research. In terms of the number of publications per
1000 researchers, as one of the key performance
measures scientific output, the Netherlands rank as the
highest in the world, already since 1994 (OECD, 2005).
With 40 biopharmaceutical publications per 1000
researchers in 1994/1995 and 49 in 1999/2000, the
Netherlands is far ahead of other European countries as
well as the US and Japan (OECD, 2005). However, the
growth of the number of publications is levelling off
and other countries are catching up quickly. Moreover,
over the period from 1994 to 2000, the focus of these
publications has shifted from fundamental research
(+3%) to applied research and technology develop-
ment (+31%). This growing shift in publications from
exploration to more exploitation oriented themes is not
backed by a particular strong position in exploitation.
According to recent figures, the Netherlands occupy,
in the overall ranking of nations, a 7th position
in terms of number of fte’s (1178 in 2000) and a
12th position in terms of numbers of biotechnology
18 A further indication for this can be found when studying the
development of the multimedia industry in the Netherlands through-
out the 1990s (see Gilsing, 2005).
20 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
companies (approximately 120 in 2000, Biopartner,
2001). So, despite its strong scientific reputation, the
Netherlands have not built up an equal reputation
in commercialisation of this knowledge (Enzing et
al., 2005). The majority of DBFs have specialised in
R&D rather than production and have not been able to
leave the early development phase. As a consequence,
most Dutch DBFs are small, employing less than 50
employees and most of them even less than 25 (Enzing
and Kern, 2002). So, in contrast to a particularly strong
position in exploration, the position in exploitation is
rather mediocre. There may be various reasons for this.
The emergence and growth of biotechnology is
shaped to an important extent by the presence and struc-
ture of downstream industries that provide demand
for biotechnology and its related products (Senker and
van Zwanenberg, 2001). Although large and integrated
pharmaceutical firms are absent in the Netherlands,
there are still a few medium-sized multinational firms
with clear involvements in pharmaceutical biotechnol-
ogy, namely AKZO Nobel (Organon, Organon Tech-
nika and Intervet), DSM-Gist Brocades (largest global
manufacturer of penicillin), Yamenouchi and Solvay
Pharmaceuticals. Moreover, foreign pharmaceutical
firms have clinical research being carried out in the
Netherlands and have relations with Dutch DBFs. So,
although the ‘demand side’ for Dutch DBFs may not
be as attractive as in some other countries, it does not
seem to form the only explanation. There may be more
reasons for this moderate strength of the Dutch phar-
maceutical biotechnology industry. In this respect, we
can differentiate between two types of factors, policy as
well as business-related. Throughout the 1980s and into
the 1990s, Dutch industrial policy has tended to favour
existing firms over new firms. Applying for R&D-funds
in this period required the ability to clearly formulate
research ideas and potential markets, which was gener-
ally easier for large(r) than for small firms (EZ, 1998,
2000). Moreover, a powerful, ethical lobby of environ-
mentalists has exerted pressure for strong regulation in
the field of biotechnology, which created another bar-
rier for start-ups (Enzing, 2003; Enzing et al., 2005).
On the business side, Dutch banks have shown to be
very risk-averse. Moreover, venture capitalists tended
to require a return on sales within 1 year, which was
far beyond the scope of start-ups with their strong
focus on R&D. Given this focus, most start-ups needed
seed capital before turning to venture capital. However,
seed capital is virtually absent in the Netherlands (EZ,
2000). Finally, another important issue is a strong lack
of a risk-taking culture and therefore virtually no incen-
tives for academics to become entrepreneur (Enzing,
2003; Enzing et al., 2005).
On the other hand, things have been changing from
around the mid 1990s. Biotechnology has increasingly
become one of the key-areas of interest for policymak-
ers. As a consequence, various policy measures have
been developed, among others the creation of a so-
called TechnoStarters Fund, aimed at raising seed capi-
tal and venture capital for biotechnology start-ups (EZ,
2000). In the meantime, from the early 1990s towards
the early years in the new millennium, the number of
entrants by DBFs per annum increased from 4 in 1994
to an average of approximately 10 in 1998 and beyond,
making approximately 120 in total in 2001 (Biopartner,
2001). The total employment in the Dutch pharmaceu-
tical industry was estimated in 2001 at 15.100 jobs and
had increased on average with 3.5% when compared
with 1994. Annual investments in pharmaceutical R&D
have also increased substantially: from D 198 million
(1994) to D 401 million (2001) (Enzing et al., 2005), an
increase of about 100%.
A.2. Some descriptive statistics on the general
profile of Dutch DBFs
In 2001, around 65% of the total firms involved
in biotechnology in the Netherlands (approximately
125 firms) were formed by DBFs with a focus phar-
maceutical biotechnology (approximately 80 in total).
These start-ups had R&D as their core activity and
engaging in R&D collaboration has become a wide-
spread phenomenon for these firms. The majority of
them (approximately 65 firms) worked closely with
universities or semi-public research institutes (Enzing
et al., 2003). Cooperation is mainly formed by means of
joint projects to which the involved parties contribute
through bringing in (a combination of) finance, spe-
cific knowledge or specialised researchers (Enzing et
al., 2003).
Most of these DBFs tended to specialise in niche
markets, niche technologies or specific activities within
the pharmaceutical R&D process, such as drug discov-
ery, lead optimisation, and/or drug delivery. In doing
so, these DBFs acted as specialised suppliers to the
large pharmaceutical firms, both inside and outside the
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Netherlands. The largest proportion of partners (35%)
with which R&D collaboration took place were based
in the Netherlands, formed by either Dutch firms or for-
eign affiliates. Next, the US formed the most important
country for cooperating with foreign partners (approx-
imately 21%), followed by the UK and Germany with
12 and 10%, respectively (Enzing et al., 2005).
These DBFs were either independently established
(60%) or spin-offs from academia (40%), with no DBFs
created through diversification from existing Dutch or
foreign pharmaceutical firms (Enzing, 2000; Enzing
and Kern, 2002). Total employment in 2001 was esti-
mated at 1764 fte’s, making an average of 23 per DBF.
Of this labour force, approximately 60% was involved
in R&D, and the total investment in R&D was almost
D 73 million in 2001 (Enzing and Kern, 2002). The
number of DBFs having successfully performed an
IPO (initial public offering) is very limited: Pharm-
ing (transgenic animals) at Nasdaq Europe in 1998
and Euronext in 1999, Fornix BioSciences at Euronext
1999, Crucell (platform technologies, gene therapy) at
Nasdaq and Euronext in 2000 and Isotis (human tissue
engineering) at Euronext 2000.
References
Abernathy, W.J., Utterback, J.M., 1978. Patterns of industrial inno-
vation. Technology Review 81 (June/July), 41–47.
Aharonson, B.S., Baum, J.A.C., Feldman, M.P., 2004. Industrial
clustering and the returns to inventive activity: Canadian biotech-
nology firms, 1991–2000, no. 04–03. DRUID Working Paper
(www.druid.dk/wp/wp.html).
Allansdottir, A., Bonnaccorsi, A., Gambardella, A., Mariani, M.,
Pammolli, F., Riccaboni, M., 2001. Innovation and Competi-
tiveness in Biotechnology: A European Perspective, Background
Report. DG-Enterprise, EU, Brussels.
Allansdottir, A., Bonnaccorsi, A., Gambardella, A., Mariani, M.,
Pammolli,F., Riccaboni, M., 2002. Innovation and competitive-
ness in biotechnology, Enterprise papers, 7-2002. DG-Enterprise,
EU, Brussels.
Argyris, C., Schön, D., 1978. Organizational Learning. Addison-
Wesley, Reading (Ma).
Arora, A., Gambardella, A., 1994. The changing technology of
technological change: general and abstract knowledge and the
division of innovative labour. Research Policy 23, 523–532.
Bettis, R.A., Prahalad, C.K., 1995. The dominant logic: retrospective
and extension. Strategic Management Journal 16, 5–14.
Biopartner, 2000. The Netherlands Life Sciences Report 2000.
Biopartner, Leiden.
Biopartner, 2001. The Netherlands Life Sciences Report 2001.
Biopartner, Leiden.
21
Biopartner, 2002. The Netherlands Life Sciences Report 2001.
Biopartner, Leiden.
Brusoni, S., Geuna, A., 2003. An international comparison of sectoral
knowledge bases: persistence and integration in the pharmaceu-
tical industry. Research Policy 32, 1897–1912.
Callon, M., 2002. From science as an economic activity to socioeco-
nomics of scientific research, the dynamics of emergent and con-
solidated techno-economic networks. In: Mirowski, P.P., Sent,
EM. (Eds.), Science Bought and Sold: Essays in the Economics
of Science. University of Chicago, Chicago, pp. 277–317.
Casper, S., Kettler, H., 2001. National institutional frameworks and
the hybridization of entrepreneurial business models: the Ger-
man and UK biotechnology sectors. Industry and Innovation 8,
5–30.
Degenaar, G.H., Janszen, F.H.A., 1996. Biotechnologie op weg naar
het jaar 2000, Een toekomstperspectief voor de Nederlandse
industrie. Erasmus Universiteit, Rotterdam.
DiMaggio, P.J., Powell, W.W., 1983. The iron cage revisited: insti-
tutional isomorphism and collective rationality in organizational
fields. American Sociological Review 48, 147–160.
Dosi, G., Marengo, L., 2000. On the tangled intercourses between
transaction cost economics and competence-based views of the
firm: a comment. In: Foss, N., Mahnke, V. (Eds.), Competence,
Governance and Entrepreneurship. Oxford University Press,
Oxford, pp. 80–92.
Edquist, C. (Ed.), 1997. Systems of Innovation, Technologies, Insti-
tutions and Organizations. Pinter, London.
Ernst, Young, 1999. European Life Sciences 1999. Sixth Annual
Report. Ernst and Young Ltd., London.
Ernst, Young, 2001a. Focus on Fundamentals, the Biotechnology
Report. Ernst and Young Ltd., London.
Ernst, Young, 2001b. Integration, European Life Sciences Report.
Ernst and Young Ltd., London.
Ernst, Young, 2001c. What’s in a Name: Leveraging a Quality Brand
in the Health Sciences Marketplace. Ernst and Young Ltd., Lon-
don.
Ernst, Young, 2002. Beyond Borders, the Global Biotechnology
Report. Ernst and Young Ltd., London.
Enzing, C.M., 2000. Bijlage bij de integrale nota biotechnologie.
Ministerie van Economische Zaken, Den Haag.
Enzing, C.M., Kern, S., 2002. The Dutch Biotechnology Innovation
System: An Inventory and Assessment of the Major Develop-
ments Since 1994. TNO-STB Report 02-04, TNO-STB, Delft.
Enzing, C.M., Kern, S., Giessen, A. v.d., 2003. R&D samenwerking
door de Nederlandse biotechnologie industrie: uitkomsten van
een survey. TNO-STB rapport 03-22, TNO-STB, Delft.
Enzing, C., 2003. Dutch cluster policy and the character of regional
concentration of biotech firms. In: Fuchs, G. (Ed.), Biotech-
nology in Comparative Perspective. Routledge, London, pp.
169–191.
Enzing, C., Kern, S., van de Giessen, A., 2005. The Dutch biophar-
maceutical and food biotechnology innovation system. OECD
Proceedings, The Biopharmaceutical Innovation System in Eight
OECD Countries, Case Studies on Structure, Performance, Inno-
vation Barriers and Drivers. OECD, Paris.
EZ, 1998. Pharmaceuticals, Technologie-radar, Ministerie van
Economische Zaken. Den Haag, pp. 65–67.
22 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
EZ, 2000. Integrale nota biotechnologie, Ministerie van Economis-
che Zaken. Den Haag.
Gambardella, A., 1992. Competitive advantage from in-house sci-
entific research: the US pharmaceutical industry in the 1980s.
Research Policy 21, 391–407.
Gambardella, A., 1995. Science and Innovation in the US Pharma-
ceutical Industry. Cambridge University Press, Cambridge.
Gambardella, A., Orsenigo, L., Pammolli, F., 2000. Global Com-
petitiveness in Pharmaceuticals: A European Perspective. DG-
Enterprise, EU, Brussels.
Gassman, O., Reepmeyer, G., von Zedtwitz, M., 2004. Leading Phar-
maceutical Innovation. Springer, Berlin.
van Geenhuizen, M., van der Knaap, B., 1997. R&D and regional
networks dynamics in Dutch pharmaceutical industry. Journal of
Economic and Social Geography 88, 307–320.
van Geenhuizen, M., 1999. How small biotechnology firms survive in
the Dutch pharmaceutical industry: an exploratory analysis. In:
Oakey, R.P. (Ed.), New Technology-Based in the 1990’s. Paul
Chapman, London, pp. 200–212.
Gilsing, V.A., 2005. The dynamics of innovation and interfirm net-
works, exploration, exploitation and co-evolution. Edward Elgar,
Cheltenham.
Gordon, E.M., Barrett, R.W., Dower, W.J., Fodor, S.P.A., Gallop, M.,
1994. Applications of combinatorial technologies to drug dis-
covery, combinatorial organic synthesis library screening state-
gies and future directions. Journal of Medicinal Chemistry 37,
1385–1401.
Henderson, R., 1994. The evolution of integrative competence, inno-
vation in cardiovascular drug discovery. Industrial and Corporate
Change 3, 607–630.
Henderson, R., Orsenigo, L., Pisano, G., 1999. The pharmaceutical
industry and the revolution in molecular biology: interactions
among scientific, institutional and organizational change. In:
Mowery, D.C., Nelsoni, R.R. (Eds.), Sources of Industrial Lead-
ership. Cambridge University Press, New York, pp. 267–311.
Knoke, D., Kuklinski, J.H., 1982. Network analysis. Sage Publica-
tions, Beverly Hills.
Jungmittag, A., Reger, G., Reiss, Th. (Eds.), 2000. Changing Innova-
tion in the Pharmaceutical Industry, Globalisation and New Ways
of Drug Development. Springer, Berlin.
Lundvall, B.-A., 1992. National Systems of Innovation. Pinter, Lon-
don.
Malerba, F., 2002. Sectoral systems of innovation and production.
Research Policy 31, 247–264.
Malerba, F., 2004. Sectoral Systems of Innovation. Cambridge Uni-
versity Press, Cambridge.
Malerba, F., Breschi, S., 1997. Sectoral innovation systems: techno-
logical regimes, Schumpetarian dynamics and spatial boundaries.
In: Edquist, C. (Ed.), Systems of Innovation, Technologies, Insti-
tutions and Organizations. Pinter, London.
March, J., 1991. Exploration and exploitation in organizational learn-
ing. Organization Science 2, 71–87.
Marsili, O., 2001. The anatomy and Evolution of Industries, Tech-
nological Change and Industrial Dynamics. Edward Elgar, Chel-
tenham.
McKelvey, M., 1997. Co-evolution in commercial genetic engineer-
ing. Industrial and Corporate Change 6, 503–532.
McKelvey, M., Orsenigo, L., 2004. Pharmaceuticals as a sectoral
innovation system. In: Malerba, F. (Ed.), Sectoral Systems of
Innovation. Cambridge University Press, Cambridge.
Mowery, D.C., Nelson, R.R. (Eds.), 1999. Sources of Industrial Lead-
ership, Studies of Seven Industries. Cambridge University Press,
New York.
Nelson, R.R. (Ed.), 1993. National Innovation Systems: A Compar-
ative Analysis. Oxford University Press, New York.
Nelson, R.R., 1994. The co-evolution of technology, industrial struc-
ture and supporting institutions. Industrial and Corporate Change
1, 47–63.
Nightingale, P., 2000. Economies of scale in experimentation: knowl-
edge and technology in pharmaceutical R&D. Industrial and
Corporate Change 9, 315–359.
Nooteboom, B., 2000. Learning and Innovation in Organizations and
Economies. Oxford University Press, Oxford (Paperback edition
published in 2001).
Nooteboom, B., 2002. Trust: Forms, Foundations, Functions, Fail-
ures and Figures. Edward Elgar, Cheltenham.
Nooteboom, B., 2004. Interfirm Collaboration, Learning and Net-
works, an Integrated Approach. Routledge, London.
OECD Proceedings, 2005. The Biopharmaceutical Innovation
System in Eight OECD Countries, Case Studies on Struc-
ture, Performance, Innovation Barriers and Drivers. OECD,
Paris.
Orsenigo, L., Pammolli, F., Riccaboni, M., Bonaccorsi, A., Turchetti,
G., 1998. The evolution of knowledge and the dynamics of an
industry network. Journal of Management and Governance 1,
147–175.
Orsenigo, L., Pammolli, F., Riccaboni, M., 2001. Technological
change and network dynamics, lessons from the pharmaceuti-
cal industry. Research Policy 30, 485–508.
Owen-Smith, J., Powell, W.W., 2004. Knowledge networks as chan-
nels and conduits: The effects of spillovers in the Boston biotech-
nology community. Organization Science 15 (1), 5–21.
Pisano, G., 2002. Pharmaceutical biotechnology. In: Nelson, R.R.,
Victor, D.G., Steil, B. (Eds.), Technological Innovation and Eco-
nomic Performance. Princeton University Press, Princeton, pp.
346–366.
Powell, W.W., Koput, K.W., Smith-Doerr, L., 1996. Inter-
organizational collaboration and the locus of innovation: net-
works of learning in biotechnology. Administrative Science
Quarterly 41, 116–145.
Reiss, Th., Reger, G., Jungmittag A. (Eds.), 2000. Changing Innova-
tion in the Pharmaceutical Industry, Globalisation and New Ways
of Drug Development. Springer, Berlin.
Roijakkers, A.H.W.M, 2003. Interfirm cooperation in high-tech
industries, a study of R&D-partnerships in pharmaceutical
biotechnology, Ph.D. thesis. MERIT, Universitaire Pers Maas-
tricht, Maastricht.
Roijakkers, N., Hagedoorn, J., van Kranenburg, M., 2005. Dual
market structures and the likelihood of repeated ties, evidence
from pharmaceutical biotechnology. Research Policy 34, 235–
245.
Santos, F.M., 2003. The co-evolution of firms and their knowledge
environment: insights from the pharmaceutical industry. Techno-
logical Forecasting and Social Change 70, 687–715.
 V. Gilsing, B. Nooteboom / Research Policy 35 (2006) 1–23
Senker, J., 1998. Biotechnology and Competitive Advantage,
Europe’s Firms and the US Challenge. Edward Elgar, Chel-
tenham.
Senker, J., van Zwanenberg, P., 2001. European Biotechnology Inno-
vation Systems, Final Report, TSER-project no. SOE1-CT98-
1117. SPRU, Sussex.
Spender, J.C., 1989. Industry Recipes. Basil Blackwell,
Oxford.
Sykes, R., 1997. The Pharmaceutical Industry in the New Millen-
nium, Capturing the Scientific Promise, Annual Lecture. Centre
for Medicines Research (CMR), London.
23
Teece, D., 1986. Profiting from technological innovation: implica-
tions for integration, collaboration, licensing and public policy.
Research Policy 15, 285–305.
Whitley, R., 1999. Divergent Capitalisms: The Social Structuring and
Change of Business Systems. Oxford University Press, Oxford.
Wuyts, S., Colombo, M., Dutta, S., Nooteboom, B. Empirical tests
of optimal cognitive distance. Journal of Economic Behaviour
and Organization, in press.
Zucker, L.G., Darby, M.R., 1997. Present at the biotechnological
revolution: transformation of technological identity for a large
incumbent pharmaceutical firm. Research Policy 26, 429–446.