We fred it hard to justify this practice of interpretation. The Method of Difference cannot ..... In: B.D. Haines and S.J. Higgins, eds.,. Transcription and Translation: ...
IN VIVO INTERPRETATION OF IN VITRO EFFECT STUDIES
with a detailed analysis of the method of in vitro transcription in isolated cell nuclei
Roger Strand*, Ragnar Fjelland** and Torgeir Flatmark* *Department of Biochemistry and Molecular Biology, University of Bergen, ~rstadvn. 19, N-5009 Bergen, Norway ** Centre for the Study of the Sciences and the Humanities, University of Bergen, Alltgt. 32, N-5020 Bergen, Norway Received 17-V-1995
ABSTRACT In vitro experimental approaches are of central importance to contemporary molecular and cellular biology and toxicology. However, the scientific value or impact of in vitro results depends on their relevance in rive. In vitro effect studies address inobservable in rive phenomena through experiments on analogous in vitro phenomena. We present a theoretical basis developed to evaluate the in rive relevance of in vitro effect studies. As a case study, the procedure for measuring specific gene transcription in isolated cell nuclei ("nuclear run-off method") is analyzed. R is concluded that current evidence fails to justify in rive interpretations of nuclear run-off experiments within the framework of theoretical models of transcription, implying that quantitative in rive interpretations are unwarranted. Qualitative interpretations of nuclear run-off experiments may be justified by inferring "the best explanation", especially when significant in vitro effects follow in rive perturbations. Elements of a general theory are proposed. R is concluded that quantitative in vivo interpretations are warranted primarily in biochemical quantitation of biomolecules, while studies on biological function should be interpreted qualitatively in terms of causal explanations. Inferences to the best explanations are strengthened through additional evidence and the creation of experimental differences (effects).
KEY WORDS: Data-interpretation; inductive logic;/n v/tro;/n rive; isolated nuclei; nuclear run-off; philosophy. Acta Biotheoretica 44: 1-21, 1996. © 1996 Kluwer Academic Publishers. Printed in the Netherlands.
1. I N T R O D U C T I O N 1.1. G e n e r a l I n t r o d u c t i o n In vitro methodologies and experiments are frequently employed in contemporary biological research, e.g. molecular and cellular biology, pharmacology, toxicology and agricultural science. The in vitro label signifies "artificial" experimental conditions, especially those of the test tube [in vitro literally meaning "in glass"], as opposed to an in vivo condition ["in the living", i.e., in the living body of an animal or a plant]. We define an in vitro effect study as a scientific study as follows:
1. The ultimate goal of an in vitro effect study is to gain knowledge of a certain biological phenomenon of a native organism or system of organisms, hereafter called the in vivo system, or Sin v/vo. 2. It is methodologically impossible, inconvenient or unethical to study the phenomenon of interest in S ~ vlvo 3. Instead, an experimental in vitro system, Sin ~ro, is produced out of one or more parts of the /n vivo system. Information on S in vivo is inferred by interpreting experimental evidence on a putatively analogous in v/tro phenomenon of S/n vitro 4. In an effect study, the biological phenomenon of interest is assumed to be an effect, i.e., a change or a difference in a biological feature or parameter following some specific perturbation of S in vivo or Sin ~ro as compared to a "control" system which was not subjected to the perturbation. S/n v/fro has to be different from S in vivo in order to overcome the in vivo methodological problems. Often, the design of Sin vitro involves the physical disruption of Sin vivo and a subsequent procedure for isolating one or more of its parts, with the aim of giving experimental access to inobservable features or parameters, or to reduce the amount of complex interactions with interfering biological phenomena. On the other hand, the in vitro experimental conditions have to be as biologically relevant as possible to allow the desired in vivo interpretation. Accordingly, researchers will normally attempt to imitate the in v/vo condition by the selection of parameters of the in vitro experiment. The scope of the present study is to address the general problem of interpreting in vitro effect studies in terms of inferences about the in vivo system. Thus, our task is to determine the conditions under which the falsification or verification of an experimentally testable in vitro null-hypothesis H j n vitro : The in vitro effect does not follow the perturbation.
justifies the corresponding falsification or verification of the related biologically relevant in vivo null-hypothesis H~n vivo : The in vivo effect does not follow the perturbation.
It might be objected that the goal of many in vitro effect studies is rather to describe the details of an in vivo phenomenon of known existence. In this case, we would define the in v/vo effect as these details. Clearly, if this descriptive study is to have any impact, it has to be verified that the observations are not due to artifacts or otherwise fictitious. This verification is equivalent with the falsification of Ho in v~vo. To our knowledge, the problem of justification of in vivo inference from in vitro
evidence is rarely discussed in biological literature. However, a similar problem has caught considerable attention in test psychology, known as the problem of "predictive validity": How can one infer future performance of subjects from results of psychological tests carded out under artificial conditions? More recently, the problem has also been attended by philosophers of science addressing laboratory experiments. Some authors have stressed the essential difficulties involved in transferdng knowledge obtained under the artificial experimental conditions to natural conditions (Hacking, 1983; Latour, 1987; Hacking, 1992). In the following section (1.2), we introduce a theoretical basis or scheme that has proved useful for evaluation o f / n vivo interpretations of particular/n vitro effect studies. Part 2 presents a case study, while part 3 discusses the prospects of a general application of the scheme. 1.2 I n t r o d u c t i o n o f a T h e o r e t i c a l B a s i s f o r E v a l u a t i n g the J u s t i f i c a t i o n o f I n V i v o I n t e r p r e t a t i o n s o f I n Vitro E f f e c t S t u d i e s
H j n vivo and H ~ vitro, defined as above, are sentences on different matters of fact, and in order to infer, say, Ho in vivo from its/n vitro counterpart, some principle of induction is required. In the present study, inferences to simple causal explanations are compared to inductive inference within model :frameworks. Inferences to simple causal explanations may be considered as a special case of the general principle of inference to the best explanation, also called hypothetical reasoning (Peirce, 1868). According to this principle, one should infer the hypothesis that, if it were true, would be the best explanation of the data. In an /n vitro effect study, one would investigate whether Ho in vivo or its negation may offer the best explanation of the observed results. This inferential strategy is particularly effective when an in vitro effect follows a perturbation of S/n vivo. If the perturbation of S/n vZvo was the only difference between the control and the perturbation experiment, the best explanation of the effect is often that the perturbation caused the in vitro effect. Inference to this kind of causal explanations is known as the Method of Difference (Mill, 1843). One might then assume that the in vivo cause was mediated to the in v/tro effect by some kind of factor that remained constant during the physical transformation of S/n vivo into S/n v/tro.If this factor can be identified with some certainty as the in vivo effect, -~Hoin ~vo may be inferred. Typically, one would look for additional evidence, preferably gained with a different experimental method, that would be explained by the postulated/n vivo effect. As pointed out by Lipton (1991), any observed fact that would be explained by the inferred hypothesis, tends to strengthen an inference to the best explanation. Simple principles as the Method of Difference are of limited value regarding scope and strength of inference. Inductive inference within model frameworks may meet a higher demand of rigor. We have adopted Giere's model framework (1991, p. 37-40), consisting of a physical system (S), a theoretical model (M), and a statement (P) claiming that system and model are similar (isomorphic) in the relevant aspects. In our case, the in vivo and/n vitro theoretical models should include all concepts explicitly or implicitly employed by their respective null-hypotheses. In addition, we introduce a "linking statement" L between the in vivo S-M-P metasystem and its in v/fro equivalent. Verification/falsification of Ho i~ v/vo is justified if the following three statements are verified (Figure 1): pin vivo. pin v~o.
Min vivo is an appropriate representation of S/n vlvo with regard to Ho in vivo. M/n v/tro is an appropriate representation of S/n v/tro with regard to H om vitro.
4 L: Various dements of M/n v/tro. (including the /n vitro effect) are true images of the corresponding elements of M m wvo (including the/n vivo effect). A comment has to be made on the definition of M/n v~o. In biological literature, "/n vitro model" usually means what we would call an S/n vitro designed to be analogous with a given S ~ vivo, i.e., a physical analogy model (National Research Council, 1985, p. 12-23). In the present study, a model is to be understood as a purely theoretical model. Furthermore, it is advantageous to define M/n v/tro in terms of our concepts of the elements of Sin v/~o rather than M/n v/vo. Thus, we separate the problem of justification of inference from experiment to /n v/tro hypothesis from our theoretical/n v/tro - / n vivo interpretational problem, as defined above (1.l). The case study (part 2) illustrates the use of simple causal explanations as compared to inference within a model framework. It is seen that a model framework analysis is preferable when effects are absent or when quantitative interpretations are sought. In the particular case study, it is concluded that currently available evidence does not justify quantitative/n vivo interpretations o f / n v/tro effect studies performed with the experimental method in question (the so-called nuclear run-off method).
Mlin vivo (theoretical model)
L: Correspondence between the elements o f
l~in vitro and M in vivo
I
of [t0 in vivo
[
(theoretical model) includes - the in vitro effect - interfering elements
includes - the in vivo effect - interfering elements
Inference in terms I
M4n vitro
pin vivo
pin vitro
s i n vivo
Inference in terms o f Ho in vitro
s i n vitro (experimental system)
II physical transformation
l,
yie, n an experimental result
Fig. 1. A Model F r a m e w o r k for the Justification of In Vivo Interpretations of In Vitro Effect
Studies. pb, ~o represents the statement "M/nvim is an appropriate representation of Si~ vi~o with regard to Ho~ ~.", while/~ v~o rep.r~ents the statement "M/nv/tro is an appropriate representation of S VI~O with regard to H om V l ~ ' O l l . See the text also (1.2).
2. A C A S E S T U D Y : TRANSCRIPTION AND TRANSCRIPTIONAL REGULATION 2.1. I n Vivo Transcriptional Regulation According to the current understanding of the Central Dogma (Crick, 1958) of molecular biology, gene expression involves a sequential flow of information from specific segments of DNA (the genes) to messenger RNA (mRNA) molecules by transcription and a subsequent translation of the coded messages into polypeptides. Associated with the specific DNA segments are various non-transcribed elements (promoters, enhancers and operators) that influence the frequency with which any transcription unit is transcribed. A number of cellular stimuli (physical or chemical) are known to induce changes in the pattern of gene expression in eukaryotic cells, as determined by a change in the cellular amount of a specific protein and its mRNA. Our case study is the investigation of effects of such cellular stimuli or perturbations upon transcription of specific genes (transcriptional regulation). At the transcriptional and post-transcriptional (but pre-translational) level, there are several possible targets of regulatory events, including the rate of transcription (Ptashne, 1988), post-transcriptional processing of pre-mRNA (hnRNA) (Chang et al., 1990), mRNA degradation (Bernstein & Ross, 1989; Hargrove & Schmidt, 1989) and the presence of naturally occurring antisense RNA transcripts (Tosic et aL, 1990). In the present work we focus on effects upon the rate of transcription, and the methodology used to determine whether such effects are present. Applying the terminology of part 1.2, the in vivo S-M-P metasystem and null hypothesis under scrutiny is
Sbt v/VOo M/n v/vo. e/n v/vo. Horn v/vo_.
A eukaryotic organism (or organ, or cell culture). A theoretical model of a cell. M/n vi~ois an appropriate representation of S/n v'vowith regard to transcriptional and post-transcriptional regulation. The rate of transcription of Gene X is equal in the cells of the control organism and the perturbed organism.
The theoretical cell of M/n vi~oincludes the following components: a nucleus, cytoplasm, substrates and enzymes necessary for transcription, processing, transport and degradation of RNA and the regulation of these processes by external stimuli. Transcription, defined as RNA synthesis by DNA-directed RNA polymerases, and RNA processing are localized to the nucleus, mRNA is exported to the cytoplasm, where it is subject to degradation. Rate of transcription of a particular gene is defined as the number of transcripts synthesized per unit time. p ~ ~/vois equivalent to a set of assumptions on in vivo transcription. The set includes the sequential flow of information already mentioned, with the additional claim that transcription (and pre-mRNA processing) is located to the nucleus, while RNA degradation occurs in the cytoplasm. It also includes an assumption of non-complexity: The/n vivo system is considered to be homogenous in the sense that one "typical" cell may represent the total cell population in terms of basal expression and transcriptional regulation, and individual cells should not interact or communicate in a way that affects the mode or rate of transcription. Interpretational problems might be present in experimental evidence (e.g. in situ hybridization and results on primary cell cultures) supporting the non-complexity assumption. However, we will assume that these problems are sufficiently small to allow
p/n v/vo to be (conditionally) verified. Unfortunately, in vivo transcription rates are not measurable by any known methodology. One is left with the alternative of /n vitro procedures for measuring specific gene transcription. 2.2. In Vitro Transcription in Isolated Nuclei Regulation of gene expression is most frequently studied by measuring the change in the steady-state level of a specific mRNA species (mRNA level) following a perturbation. Experimentally, RNA is isolated from a tissue or cell population of the perturbed and the control in vivo organism. Specific mRNA species is then detected by their ability to hybridize with labelled eDNA (or antisense) probes or to facilitate translation of a specific protein in an/n vitro translation system. Alternatively, tissue from the/n vivo organism is sectioned and mRNA:labelled eDNA hybridization is performed in s/tu. The regulatory mechanism causing a change in the steady-state mRNA level is experimentally more difficult to define. The method of in v/tro transcription in isolated nuclei (nuclear run-off; also called nuclear run-on) aims at providing information on the mechanism of transcriptional/post-transcriptional regulation (Marzluff & Huang, 1984). The essential stages of an in vitro transcription assay are the following: Cell nuclei are isolated from the tissue or cell population of interest. The purified nuclei are incubated at high concentrations of the ribonucleoside triphosphates (ATP, CTP, GTP and UTP), one of which (normally UTP) is labelled (with a radioactive, e.g. 32p, or chemiluminescent chromofore) in order to label nascent RNA transcripts. The nuclear RNA is then isolated and hybridized to a eDNA or antisense probe (spotted onto a suitable membrane) corresponding to the gene of interest, and the label is quantified. The resulting data are often presented as scintillation counts of (32P-labelled) RNA-hybridized slot or dot blots or densitometry areas of autoradiographed blots. As the data are relative numbers, they are often presented as "fold increase", i.e., the perturbed:control ratio. Once being considered a technique employed by the utmost experts only, the nuclear run-off method was applied in more than 400 original articles published 1993-1994, according to data of the Medline® database. Applying the terminology of part 1, the in v/fro S-M-P metasystem and null hypothesis under scrutiny is S~ ~'ro.
the experimental system consisting of isolated nuclei, reagents (including ribonucleoside triphosphates) and a detection system for/n v/tro transcription. M/n v/~o: a theoretical model of the experimental system. p/n v/~ro: M/n ~ro is an appropriate representation of S/n v/~ with regard to nuclear runoff experiments. Hjn rico: The in vitro transcription produced an equal number of nascent RNA transcripts complementary to cDNA/antisense probes of Gene X in the cell nuclei isolated from the control and the perturbed S/= ~/vo. The theoretical model, M/n ~ro, includes nuclei and an incubation medium with relevant reagents, in which the nuclei are capable of in vitro transcription. In vitro transcription is defined in terms of nuclear run-off data obtained with the desired eDNA or antisense probe, caused by a DNA-directed incorporation of ribonueleosides into nuclear RNA. Other processes are not included in the model. The justification of pin ~n,, and thus the verification or falsification of Ho/n ~ro, demands methodological evidence. First, it should be verified that there is no degradation
of RIgA in S/n v~ro. Moreover, it should be shown that the hybridization data are due to specific recognition of the cDNA/antisense probe, which may be a more difficult task. Often, data are compared with "reference probes", i.e., DNA probes that are not supposed to be complementary of any strongly expressed transcript, and for that reason may be used for estimating the degree of non-specific binding (see also 2.3.1 and 2.3.2). However, reference probe data merely indicate the degree of non-specific binding of the actual "G-erie X" probe, a problem which is seen in any hybridization technique. The typical methodological answer is variation of stringency of hybridization washes until maximum difference of perturbed and control is obtained. This strategy is somewhat problematic in a nuclear run-off experiment, both because signals are often low and hard to detect, and high stringency reduces the specific signal to some extent, and because one is trying to detect an RNA population with possibly many small fragments. Accordingly, a high stringency may lead to difficulties in telling whether there is more (in mass) of a particular labelled mRNA species, or there is merely a change in size distribution. Methodological control experiments designed to investigate such questions are rarely seen in the literature. We conclude that the justification of p/n ~ro is not trivial, and demands methodological controls that are more extensive than what is current scientific practice. 2.3. I n f e r e n c e f r o m I n Vitro Transcription to In Vivo Transcription In our experience, almost every scientific report presenting nuclear run-off data contains an in vivo interpretation of the data. However, it is not unusual that analogous data are interpreted differently by different authors. In part 2.3.1, the justification of qualitative interpretations (equivalent with a falsification or verification of Ho/n v/vo) is analyzed. Quantitative interpretations compare the "fold increase/decrease" or "fold difference" in steady-state m_RNA levels and in in vitro transcription in order to decide upon both the existence of transcriptional regulation and the presence of post-transcriptional regulatory mechanisms. This type of inference is the topic of part 2.3.2. 2.3.1. Qualitative Interpretations w e. , . in vitro present, -Ho may be scen to be the best explanation under the assumption of pm vJvo and partially P~ v ~ r o , and with some additional evidence, as follows: Assume that the outcome of a nuclear run-off experiment was R+:
R+: The nuclear run-off result with a probe of Gene X was significantly different in the perturbed and the control parallel of S/n ~ro. Assume that R+ appears reproducibly in a particular (repeated) experiment. Next, assume that the experiment was properly performed in the sense that the /n vitro samples of the perturbed and control parallels were manipulated identically in every relevant way, that the perturbation was the only systematic difference in the/n vivo organisms, and that the nuclei were purified by an optimal procedure. The best explanation, by Mill's Method of Difference, would clearly be: R+ was caused by the in vivo perturbation. The perturbation caused an effect in the nuclei which caused R+. We have shown that it is no trivial task to provide a full justification of p/n v/fro (2.2). However, assume that R+ does not appear in so-called negative controls, as when
8 ribonucleoside triphosphates are not present in ~ ~ ' , or if an inhibitor of RNA polymerase (e.g. a-amanitin for eukaryotic RNA polymerase H) is present, or if the samples are incubated with high amounts of RNase after the nuclear run-off assay. Recalling that the best explanation ought to be likely, with explanatory power and without unnecessary complexity (Lipton, 1991), it may be inferred that:
The perturbation caused an effect on the nuclei which resulted in a difference in the in vitro transcription of nuclei from the perturbed and the control parallels. One would assume that the effect on the nuclei was mediated by some material, structural or functional factor(s) (1.2). We do not know the in vivo function(s) of the factor(s). However, if evidence for a regulation of expression is present (e.g. a similar change in steady-state mRNA level upon the perturbation), the following explanation is able to account for both this evidence and R+ (together with its explanations as stated above):
-~H~ vivo: The rate of transcription of Gene X was different in the cells of the control organism and the perturbed organi.vm. In other words, the regulatory factor(s) responsible for the difference in in vitro transcription and in the steady-state mRNA level, worked in the same way in the in vivo system. When an in vitro effect is not present, the method of inference to the best explanation is less powerful. Assume that the outcome of a nuclear run-off experiment was R : R:
The nuclear run-off result with a probe of Gene X was not significantly different in the perturbed and the control parallel of S~ v/tro
The reasoning from R+ to -~H~ vivo rested upon an observed difference, which was found to have an in vivo cause. In the case of R a similar argument is not supported, because one is confronted with two very likely explanations:
1. Hom vitr° andHo ~v/v° or 2. The physical and chemical manipulations of S ~ ~ivoand S in vitro introduced HnoiseWso that the putative effect of the perturbation could not be detected. Various control experiments may indicate that the latter explanation is implausible. First, one ought to show that one is capable of detecting transcriptional regulation of the gene in question with a particular experimental set-up. Thus, one should present, together with R_, a R'+ which was obtained in a simultaneous and identical experiment except for the choice of in vivo perturbation. (Checking the probes with various amounts of total labelincorporated RNA is not sufficient, as one might be detecting some '"oackground" signal which is not reflected in the reference probes.) Furthermore, a second parallel experiment should show an R"+ with a different probe, but the same label-incorporated RNA which yielded R , to demonstrate the quality of the in vitro transcription. However, even if these parallel data are given, it is in our view still not clear beyond doubt that H0/~ v/voreally is a better explanation than the alternative "noise" explanation.
t3g. 7. Opposite page. A Model Framework for the Justification of In Vivo Interpretations of Nuclear Run-Off Experiments. It is concluded in the text that the assumptions/~ ~/~o and p ~ ~/~,o can be seen as valid (2.1 and 2.2, respectively), while L does not hold (2.3.1).
sin vivo
in vivo perturbation vs control
-
Isolation of cell nuclei from rat tissue
II
sin vitro
Different or equal amount of radioactive material bound to membrane where a DNA probe was spotted.
Experimental result:
Nuclear run-off experiment with "perturbed" and "control "nuclei.
~
-
Inference presupposes pin vitro: no degradation in S~nvitro - hybridization due to specific recognition of the probe
Different amounts of nascent RNA transcripts RNA for hybrid/zatlon complementary withprobe0I~n, x to the probe.
L: Nuclei, DNA, RNA and transcription in vitro are true images of their in vivo counterparts.
In vitro effect:
Different rate of transcription in cells of perturbed and control animal.
i ~ n vitro
In vivo effect:
Inference presupposes pin vivo: - the "Central Dogma" assumptions upon localization of processes - non-complexity assumptions
and degradation
c~oplJsm
i ~ n vivo
10 Analysis within the model framework (Figure 2) described above (1.2; 2.1; 2.2) is in principle powerful enough to warrant inference of rio~ ~ when an effect is absent, pin and p/a ~ro was found above (2.1; 2.2) to be verifiable with some certainty. We proceed to investigate whether L, defined as below, may be verified by currently available evidence. L: The various elements (nucleus, DNA, transcription, mRNA) of M/~ ~ro are true images of the corresponding elements of M/n ~o It will be sufficient to discuss whether /n v/tro nuclei and transcription are the true images of their/n v/vo counterpart. Nuclei isolated under optimum conditions clearly share morphological features with nuclei within intact cells (Blobel & Potter, 1966). However, similar rates of /n vitro transcription have been observed in experiments on heavily aggregated/disintegrated nuclei and equivalent samples of apparently intact nuclei (Strand, unpublished observations). When the structure of/n ~tro transcribed RNA is to be studied, chromatin extracted from disrupted nuclei is the substrate of choice (Stewart Gilmour, 1984). In vitro transcription is different from/n v/iv transcription in several respects. In vitro transcription stops after approx. 15 rain. of incubation, while/n v/vo transcription is a continuous process (although with large diurnal variations). The/n vitro transcription assay in isolated nuclei was termed "nuclear run-off' as one explained the gradual cease of transcription with the idea of the RNA polymerase complexes "running off' their templates, with a low ability of new initiation after termination. This difference is nevertheless unimportant if transcripts initiated /n v/vo are faithfully elongated /n ~qtro. However, Marzluff & Huang (1984) stressed the following point: "The rate of RNA synthesis [...] represents only 5-10% of the/n vivo rate. On the other hand, the rate of elongation/n vitro (5 nucleotides/sec) is 20~ of the estimated/n vivo rate of 25 nucleotides/sec. Thus it is not obvious that all RNA polymerase molecules which initiated their RNA transcripts/n v/To simultaneously elongate these/n ~ro." L is clearly not justified if the composition of transcripts from/n v/to and /n vitro transcription is different. One should note the following: (1) A nuclear run-off experiment yields only a relative number intended to be proportional with the amount of labelled specific RNA species synthesized during the assay divided by the total incorporation of label into RNA. This standardization is commonly used because the total incorporation of label into RNA tends to vary, even in "identical" parallels (the assay is sensitive towards mechanical manipulations, e.g., mixing procedures (Strand et aL, 1994)). Thus, one must measure the (/n vitro) total RNA synthesis to know whether the perturbation had any effect on the total rate of transcription. (2) The relative contributions from RNA polymerases I, II and HI might be regulated by the perturbation. In addition, it is known that the three polymerases have different salt optima (Jacob & Rose, 1977), which calls for attention to the composition of the nuclear resuspension medium/n vitro. (3) One cannot be sure that the specificity of elongation is the same/n vitro as/n vivo. The use of glycerol in nuclear resuspension media illustrates this problem. 10-40% (vN) glycerol is nearly always used to preserve morphology and RNA polymerase activity in freeze-stored nuclear preparations. Skripal & Babichev (1989) reported of a "hydrophobic effect" of glycerol (and dimethylsulfoxide) o n / n v/tro transcription. They observed that a preincubation of DNA with 10% (vN) glycerol or 5% (vN) dimethylsulfoxide increased by 60-90% the subsequent synthesis of total RNA by added prokaryotic RNA polymerase, whereas a similar preincubation of the enzyme had no effect. Conformational changes in DNA (due to
11 hydrophobicity of the surrounding medium) that might affect local capacities as transcription templates may explain both this observation and reports of extreme salt optima of RNA polymerase (Widnell &Tata, 1966; Jacob & Rose, 1977). It is reported (Strand et al., 1994) that salt and glycerol concentrations of the nuclear resuspension medium affects the mode of transcription: In a hyperosmotic glycerol medium the overall/n vitro transcription was high, while the relative (standardized) signals of two specific gene probes were low. In an approximately isoosmotic sucrose medium the overall transcription was somewhat lower, while the relative signals of the same probes were twice as high as with the hyperosmotic glycerol medium. (4) Some transcription studies have been interpreted to show the existence in vivo of transcriptional control mechanisms involving paused transcription complexes previous to or early in elongation (Rougvie & Lis, 1988; Strobl & Eick, 1992; Eick et al, 1994; Schilling & Farnham, 1994). The same studies indicate that the paused transcription complexes are released during the isolation of nuclei, giving rise to "inappropfiete" transcription in vitro. At present, it is known how common this "hold back" phenomenon might be in vivo. There are other questions about the relation between/n vivo and in vitro transcription that hitherto have remained unanswered (if not also unattended). The mechanism of transcriptional regulation is generally considered to involve regulatory DNA-binding proteins (transcription factors). Are such proteins retained in a native form dating the nuclear isolation procedure, which often includes mechanical homogenization, detergents and/or hyperosmotic centfifugation gradients? The in vitro transcription incubation is typically performed in a medium containing millimolar concentrations of ribonucleoside triphosphates in order to produce enough labelled RNA. Do these non-physiologically high concentrations influence the specificity of RNA polymerases? It can be concluded that there is some evidence against L, while other aspects of it remain unanswered. Accordingly, L is at present not to be accepted, and inference from in vitro to in v/vo within the proposed model framework is not justified.
2.3.2. Quantitative Interpretations Comparisons of results of/n v/tro transcription experiments in isolated cell nuclei with steady-state mRNA levels are frequently interpreted as to indicate the presence or absence of additional post-transcriptional regulatory mechanisms, and selective mRNA stabilization/destabilization in particular. Such interpretations are not only given when nuclear run-off data are "negative", but also when the fold increase/decrease in in v/tro transcription is smaller in magnitude than the corresponding change in steady-state mRHA levels. The latter type of interpretation, which was found in more than 50 full-length articles published in international journals in 1993 and 1994 (references not shown; see also below), rests upon several additional assumptions. First, it assumes the following relation between the cellular amount of an mRNA species (mp~A), the rate of transcription (ks) and the rate of degradation (kd) (Hargrove & Schmidt, 1989):
dme,NA dt
ks
k dml~vA
(1)
Thus, transcription kinetics are assumed to be of zero-order, while degradation is considered to follow pseudo-first-order kinetics. Furthermore, one assumes that the initial amount mRNA is equal in the control and the perturbed-to-be organism previous to the perturbation, and that the rates of transcription and degradation are constant except from discrete changes at
12 the onset of the perturbation. At a given time t after the onset, the ratio between the perturbed (p) and control (c) with respect to the amount of a specific mRNA is given by eq. 2:
k:_ kp
m
m(Of
ky
¢
k;
m(t) RNA
'
k•
"~c - i
-
mRNA
e-kf,
/
(2)
initlal e-k:t m1~VA
In the case where the rate of degradation is equal in the perturbed and the control, a rearrangement of eq. 2 yields eq. 3: . .p initial -kd k ; -- m(t)RNA-mRNA e (3) . - c
initial
m(t)RNA-mRNA e
-k d
which at steady-state conditions reduces to eq. 4: dmRNA - 0 =~ ksp - ml~VA
dt
k;
m;A
(4)
Equations 3 and 4 imply that upon transcriptional regulation, a change in/n vivo amounts of a specific mRNA will be accompanied by an equal or larger change in /n v/vo transcription rates. Employing the view that/n vivo mRNA levels and transcription rates are measured in/n v/tro hybridization of total mRNA and transcription in isolated cell nuclei, respectively, smaller changes in /n v/fro transcription than in mRNA levels have been interpreted as evidence of additional regulatory mechanisms. Examples taken from international journals of the year 1993 are Bj6rkhem et aL, Brown et al., Clarke et aL, Cochary et aL, Davis & Felder, Hoekraan et aL, Marie et aL, Salehi-Ashtiani & Goldberg, Speth & Oberbaurner, and Timmers et aL; equivalent examples of 1994 include Dchio & Schell, Eader et aL, Kubicki et al., I~n et al., Tam & Deeley, Yount et aL and Zhou et al. We fred it hard to justify this practice of interpretation. The Method of Difference cannot be applied to unequal magnitudes of effects, and inference within the model framework of part 2.3.1 was found to be unjustified as L could not be verified. Another possibility for justification we may call "the noise argument": The in vitro effect is a true (i.e., not distorted) image of the/n vivo effect except from the addition of a noise element, due to unspecific hybridization, occasional damage to biological and chemical structures, suboptimal /n vitro conditions, etc. However, there are serious objections to the noise argument. First, as discussed above (2.3.1), the estimation of the "background" contribution is non-trivial. Secondly, it is not easy to verify the assumption that the "true"/n vivo signal is not distorted while transferred to the/n vitro system. The objections against L would also apply in this case. For instance, if the nuclear isolation procedure caused a signal amplification, the real nature of this artifact would be practically impossible to discover experimentally: One can hardly exclude the possibility that e.g. nuclear transcription factors are affected by the nuclear isolation procedure. They might be washed out to some extent, or dislocated, or experience structural and functional
13 perturbations. If the noise argument, and eq. 3-4, is to apply, such changes would have to obey pseudo-first-order kinetics. What reason is there to assume that completely unknown processes would obey pseudo-first-order kinetics? Similar arguments on the other putative "noise processes" (e.g. efficiency of hybridization as a function of size distribution o f / n vitro transcribed fragments) can be produced. Thus, the noise argument is not yet justified in the case of in vitro transcription. In summary, it can be concluded that quantitative in vivo interpretations of in v/tro transcription (nuclear run-off) experiments are not justified. '2.4. I m p l i c a t i o n s We concluded above (2.3) that/n vivo transcriptional regulation may be inferred from "positive" nuclear run-off data, that the/n vivo relevance of "negative" nuclear run-off data is problematic, and that comparisons between nuclear run-off data and steady-state m R N A levels do not allow conclusions on post-transcriptional regulation/n vivo. A n immediate implication is the criticism of several authors in the past (2.3.2) and a call for a more careful practice of data interpretation. The analysis has shown that the question "Is there transcriptional regulation?" can be answered "Yes" or "It seems not". Rather than confirming "negative" results through additional control experiments (e.g. pulse labelling of RNA or reporter gene assays, as proposed by Schilling & Famham (1994)), it is preferable to detect the presence of some other mechanism, e.g. regulation of mRNA stability. There seems to be an increased tendency to supply such evidence through so-called mRNA clearance assays, in which cellular m R N A levels are measured as a function of time after the addition of a potent inhibitor of RNA synthesis, actinomycin D in particular. However, if one detects a difference in m R N A half-lives, the conclusion ought to be "There is regulation of m_RNA stability" and not "There is no transcriptional regulation". "How much do transcriptional regulation, selective mRNA (de-)stabilization and other post-transcriptional regulatory processes contribute to the overall gene regulation/n vivo? What is the major level of regulation?" are central questions within current research on mechanism of gene regulation. Unfortunately, the scientific ambitions of providing adequate and justified answers to these questions cannot be met by the present level of experimental sophistication. At present, alternative experimental approaches are scarce (reporter gene assays and techniques to study structure and amount of cellular RNA under non-steady-state conditions, e.g. pulse labelling). Although new methods may appear, any experimental approach will involve its particular problems of interpretation. For instance, the in vivo relevance of reporter gene assays depends upon the assumption that the structure of the RNA-encoding region of a gene, as well as its chromosomal position, is irrelevant to transcriptional gene regulation. (See also 3.1 for a discussion of cell culture studies.) It is therefore interesting to discuss the hypothetical case in which the nuclear run-off method will not be superseded by new and better methods. Three possible future scenarios may then be suggested: (1) A shift of focus from realistic interpretations of nuclear run-off experiments (asking for the in vivo truth) to instrumental interpretations (asking for the regulatory mechanism most suitable for manipulation through drugs etc.). (2) Success of quantitative, realistic interpretations: Through refinement of technique, changes in steadystate m R N A levels might be explained quantitatively by the contributions from the various regulatory mechanisms, the best explanation of which is that the assays somehow reflect
14 the true in vivo phenomena. (3) Searching for the major level of gene regulation will be regarded as a degenerating research programme (Lakatos, 1970): Due to the biological complexity of the cell and its massive causal interrelations, one may never establish a true understanding of gene regulation. If the research community comes to appreciate the problems of complexity, the interest in intracellular causal relations and thus the identification of regulatory mechanisms will disappear. Belief in these scenarios obviously has implications for the practical question of how to study gene regulation. It is equally obvious that this issue is not settled yet.
3. E L E M E N T S O F A G E N E R A L T H E O R Y O N I N V I V O INTERPRETATION OF IN VITRO EFFECT STUDIES To the authors' knowledge, there is no previous general theory on the justification of in vivo interpretations of in vitro experiments. In this paragraph, we therefore discuss the relevance of the proposed scheme for the analysis of the problem of justification, as presented above (1.1). Scope of the presented approach. The definition of "in vitro effect study" (1.1) excludes research directed only at optimizing in v/tro bioproduction or predictive abilities. An example of the latter is "phenomenologicar' in vivo-in vitro correlation studies, as seen in various biomedical disciplines, characterized by the possibility of obtaining in vivo data, although in limited numbers, combined with a downplay of interest in mechanism. The problem of justification of in vivo interpretations in such studies has been addressed by so-called validation studies (Gad, 1993) and some theoretical discussions (Scala, 1987; BSAC Working Party, 1991). A radical criticism of pharmacokinetic in vivo-in vitro correlation was put forth by Hiittentauch & Speiser (1985). Interpretational problems in biomechanical testing were reviewed by Kostuik & Smith (1991). Formation of in vivo concepts and theories of complex biological phenomena at the cellular/subcellular level (intracellular transport, hormone action and cellular signal transduction etc) seems to be based upon complex processes, in which in vivo theories gradually emerge from reviews of large sets of experimental data obtained with various in vitro methodologies. The approach of the present study is not intended to explain the formation of such "grand theories". Rather, our approach is more localized, investigating simple inferences between analogous phenomena in vitro and in vivo. One may object that in vivo interpretations are unnecessary at the level of a particular experiment, but should await the formation of a "grand theory". It is our view that this objection is not valid. The in vitro data that are considered when a "grand theory" is formed should be biologically relevant, and the justified possibility of a qualitative in vivo interpretation of the data would seem to be the proper choice of criterion of biological relevance. Besides, the localized in vivo interpretation is the more important one when the perturbation is the main object of study, or when medical, ecological or experimental decisions are to be made. A ubiquitous example of the latter is the decision on what experiment to do next. 3.1. I n Vivo I n t e r p r e t a t i o n w i t h i n a M o d e l F r a m e w o r k
In the following, we analyze the prospects of justification through a model framework according to the nature of the in vivo effect: whether it is related to function or the mere physical presence of biological entities.
15 Model frameworks, biological function and in vivo complexity. In order to establish the necessary correspondence between in vivo and in vitro (S-M-P) metasystems (1.2) in the case of a putative effect on a biological function, one has to know every factor that interferes with the biological function of interest, and what effect it has in vivo and in vitro. I f these factors also have to be studied by in vitro experiments that in turn are interpreted in v/vo, new justification problems will occur, and care must be taken to avoid circularity or a vicious regress in the argument. An important example of in vitro effect studies on biological function are stimulusresponse studies on cell cultures. Hirschelmann (1991) pointed out that there are several biologically important differences between a cell culture and its in vivo origin: The cell culture lacks the homeostatic potency of the organism, which may give rise to "false" effects in vitro; the in vivo --, in vitro transformation may introduce noise and erase true in v/vo phenomena; "holistic" features as chronobiological influences and undulating plasma concentrations are lacking in vitro. These objections, together with severe in vivo - in vitro discrepancies in research on anti-inflammatory agents, led to speculation upon a "oiological uncertainty principle" (Hirschelmann et al., 1990). Most of Hirschelmann's objections are due to the fact that organisms, organs and tissues are systems of higher complexity than cell cultures with respect to biological phenomena. Now, the construction of an in vitro method (when the in vivo effect is not directly observable) presupposes that the in v/vo effect is located to one part of the in vivo system, and that this part may be isolated. For instance, we assumed in our case study that transcription is confined to the cell nuclei. However, although a biological process may be precisely located with regard to mass and energy, it is more difficult to show (and less likely) that it is closed with respect to information. For instance, it is common to observe expressional changes (dedifferentation) of tissue-isolated and cultured cells, although the culture media are designed to imitate the in vivo tissue fluids (Reid & Jefferson, 1984). Furthermore, theoretical exploration of dynamic (mathematical) models has shown that even intracellular regulatory mechanisms of sufficient complexity can exhibit an extremely high sensitivity (Swillens & Pirson, 1994), and it is an open question whether the native level of sensitivity can be retained in a cultured cell. Control experiments for the detection of complexity with respect to the justification of in vivo interpretation would be seen to compare S/n vlvo with a parallel that is held to be less complex with respect to the in v/tro effect in question. Two examples are given for illustration. At the molecular level, possibly interfering intra- and intercellular processes may be blocked by the addition of relatively specific irthibitors as cycloheximide. In vivo parallels with reduced complexity at a macro-level may be of great scientific importance, not only to detect in vivo complexity, but also in their own right as in vitro systems. An interesting example is the method of making "minibrains", i.e., cultures of isolated and reaggregated fetal brain cells (Bjerkvig et al., 1986). There are essential difficulties attached to the problem of detecting complexity through control experiments. First, the difference between the in vivo system and the hypothesized non-complex parallel will be monitored by some (typically in vitro) method that might be unable to detect the difference in complexity. Secondly, the belief that a non-complex parallel is produced ought to be justified. If firm in vivo evidence is not available, the situation might be exactly that of a regress or circle of interpretational problems, and simpler inductive approaches would be preferred to an unsuccessful justification within a model fxamework.
16 Model frameworks and the physical presence and quantity of biological entities. The argument on complexity, as stated above, applies in principle to studies on the physical presence and quantity of biological entities. For instance, the amount of a biomolecule is frequently determined by monitoring a function that is considered to be invariably linked to the molecule. If the nature of this correlation depends on the physical environment, one also has to justify t h e / n vivo - in vitro correspondence of the experimentally underlying biological function. However, one may counter the complexity arguments with the presumably reasonable opinion that the amount of molecules, cells and organisms etc. does not change in the absence of an active cause. It follows that quantitative in vivo interpretations may be justified by verifying L in a "negative" sense: The relevant amounts cannot have changed significantly during the in vivo --~ in vitro transformation and the subsequent in v/tro measurement. Thus, the kinetic aspect (in terms of stability, time and temperature) is frequently more important in biochemical methods for quantitation than the imitation of in vivo conditions. An example is the isolation of RNA by extraction in strongly chaotropic (protein denaturing) solutions (Chomczynski & Sacchi, 1987). Conclusion on model frameworks. A general hypothesis concerning the justification problem of inference from in vivo t o / n vitro may then be proposed: A strong justification of in v/vo interpretations, e.g. through a model framework, would a priori be preferred to simple causal explanations (1.2; 2.3.1-2). Model frameworks may succeed with studies on the number or amount of biological entities and with studies on a biological function about which there is a large body of knowledge present. However, the remaining part, novel functional discoveries, is an important part of biology. It follows that there is a pressing need for the more simple inductive approaches to justification.
3.2. I n V i v o C a u s a l E x p l a n a t i o n s The use of a difference in causal explanations. The present work stresses the value of subjecting the biological system to an external perturbation in order to create an effect. While some philosophers of science (Popper, 1980, p. 106-111) have stressed the need of theoretical hypotheses previous to scientific discoveries (so that one would know what to look for), it is a fact that several discoveries have resulted from observations of perturbed systems. Thus, the function of the cytoplasmic organelle called the peroxisome, although ubiquitous in eukaryotic cells, was not discovered until rats were fed with ethyl-ctchlorophenoxyisobutyrate (clofibrate), a compound now known to cause peroxisome proliferation and changes in the cellular lipid metabolism (Paget, 1963). A more famous example would be the discovery of the antibiotic effect of penicillin (Fleming, 1929). I n vivo vs in vitro perturbations. In certain types o f / n vitro effect studies one has the opportunity to choose between a perturbation of the in vivo or the in vitro system. For instance, one may administer a drug to an animal and then perform in vitro studies on a primary cell culture from one of its tissues, or one may choose to obtain a cell culture from an "unperturbed" source and then subject this cell culture to the drug in vitro. Present research displays a preference towards in vitro perturbations in such cases, presumably for practical reasons mostly. However, from a theoretical point of view, in vivo perturbations are preferable to in vitro perturbations: In the former case the Method of Difference is applicable, while in the latter ease one has to assume that the effects are identical or similar whether the perturbation is performed /n vivo or in vitro. We would therefore like to encourage the use o f / n v/vo perturbations.
17 The use of additional evidence in inferences to the best explanation. Studies on cells and tissue do have one significant advantage over say, our case study of in vitro transcription in isolated nuclei. When applying the experimental method of our case study, additional evidence has to be provided to verify the qualitative conclusion of the experiment and even to claim that the experiment was successfully performed (2.2; 2.3.1). In contrast, when the experimental system includes calls or tissue, morphological techniques as microscopy and in situ hybridization supply large amounts of redundant information, which may include other in vitro effects that would be explained in terms of the in v/vo nullhypothesis, or detect experimental failure, or serve as to evaluate the general similarity between the in vivo and the in vitro system. Furthermore, additional evidence in e.g. a morphological study is typically correlated intrinsically and to the main data ("multivariate"), and thus contains more information than the same amount of analogous evidence obtained by independent ("univariate") measurements.
4. D I S C U S S I O N Since the first experiments on tissue culture early in the 20th century, in vitro methodologies undoubtedly have revolutionized biology and medicine. In the early 1990s, tens of thousands of in vitro studies were published every year, according to the Medline ® database. Likewise, hundreds of articles and books have treated methodological questions of in v/tro research. However, the focus of interest has been almost exclusively on the practical issues of designing, choosing and improving in vitro systems, even in the more theoretical studies (National Research Council, 1985). In this respect, the present study is unusual. This discussion is an attempt to clarify the purpose, methodology and limitations of the theorizing approach of this work. Purpose and methodology. The proposed theorization of interpretational practices in in vitro effect studies is intended to be useful and empirically adequate rather than "true" or "approximately true". The purpose of the study was to find a (not the) conceptualization and verbalization that could deal efficiently with the interpretational problems involved. A suitable methodology for this purpose was found to be that of the philosophy of science: descriptive representation and reinterpretation of selected scientific achievements and practices, together with normative speculation through philosophical argument. Any particular methodology introduces a set of limitations. The use of a case study is discussed below. Furthermore, we have had to exclude some aspects of in vitro biology (see part 3) and simplify certain concepts, in order to obtain a uniform theoretical basis. Thus, we have not distinguished between an experiment and a method, or between assertions and weaker claims of knowledge. The relevance of these distinctions is discussed below, together with our possibly simplistic categorization of in vivo and in vitro systems. The particular case study is chosen partly out of didactic reasons and undoubtedly as a result of the author's idiosyncrasies (Strand et aL, 1994). However, the case study may also be significant as more than an illusttation, especially in the argument on model frameworks and biological functions (3.1). There has been an immense scientific attention to the processes of gene expression since the discovery of the double helix structure of DNA. It is recalled that justification of in vivo interpretations through a model framework was unsuccessful in the case study (2.3.1), partly because of the lack of sufficient knowledge about the transcription process and phenomena that interferes with it. One could
18 ask rhetorically: Of what biological intra- or intercellular function do we have more knowledge than that of the gene expression apparatus? Will any such biological function be able to pass the test of the model framework? Of course, this is a suggestive type of argument without any formal validity. Experiment vs method. The distinction between a particular experiment and its method was ignored in this study. This distinction is irrelevant when discussing well-defined methods (e.g. statistical methods as Student's T test), but ought to be treated with care in biological experimental disciplines. For instance, different research groups are normally found to conduct the nuclear run-off assay (see part 2) differently with respect to the actual experimental details. Besides, even if assay conditions are equal in terms of concentrations of chemicals etc., an assay will frequently '"oehave differently" when applied to largely different biological systems (e.g. different cell types or organisms). One might speculate whether it is common to adapt in vitro assay conditions according to the biological system of study, in which case the "method" in use can hardly be considered as a well-defined entity. On the other hand, progress in science seems impossible without allowing generalization between similar experiments, i.e., experiments whose similarity constitutes a "method". From a pragmatic viewpoint, one will have to accept reasoning by analogy between different studies performed under similar conditions. However, one should be aware of the qualitative difference between accepting the pragmatic argument above and allowing inferences from in vitro to in vivo based upon a general in vivo - in vitro "similarity". The latter practice seems to be unfounded both theoretically and pragmatically. Assertions vs indications. In the present work, interpretational sentences are stated as assertions, in terms of falsification or verification of hypotheses. It might be objected that truly scientific statements are contingent and merely indicate that the hypothesis would be true or false if certain assumptions were shown to be true, implying that our analysis of assertoric interpretational sentences is irrelevant. However, it is our view that this objection is not valid, although it is true that statements of the type "The data indicates that..." are more commonly found in scientific literature than "The data verify/falsify the hypothesis on..." In order to build a body of actual scientific knowledge which can be applied to the surrounding world, scientific statements have to be decided upon: At least they must be regarded as candidates for temporary verification or falsification. We have intended to analyze this decision. Such analyses may indicate what future methodological evidence is needed for a justification of inference, or it may even conclude that the /n vitro system under study is badly chosen. I n vivo vs in vitro. The categorization into/n vivo and/n vitro systems is to some extent associated with the concepts of the natural and the artificial, but this analogy is not without problem. Transgenic mice is certainly considered to be in vivo systems, although highly artificial. Natural biological variation is frequently reduced in the/n vivo system of study by choosing genetically homogenous organisms (e.g. laboratory strains of rodents) and, for instance, denying laboratory animals access to food the day previous to their use in in vitro experiments to suppress chronobiological variation (Barbiroli et al., 1973). Although in vivo inferences - in terms of the biology of laboratory organisms - are provided more easily, it is evident that a new justification problem arises from the described laboratory practices: Is knowledge about laboratory organisms relevant for the biological understanding of natural organisms? F u r t h e r developments. If one accepts the relevance of the approach and the arguments presented in this work, one will find that several questions should be attended. First, the
19 presented theoretical basis should be applied to various in vitro methodologies in order to clarify their in v/vo relevance. Next, theoretical bases should be developed for the study of in vivo - in vitro correlation disciplines and the formation of "grand theories" from arrays of in vitro data (part 3). Finally, a theoretical analysis of the distinction between experiment and experimental method would be welcomed.
ACKNOWLEDGEMENTS O. Storeb~ is acknowledged for his advice on description of inductive practices. R. Hirschelmann, K.E. Fladmark and S.A. Sande are acknowledged for their help during the literature study. This work has been supported by the Research Council of Norway and the Norwegian Cancer Society.
REFERENCES Barbiroli, B., M.S. Moruzzi, M.G. Monti and B. Tadolini (1973). Diurnal rhythmicity of mammalian DNA-dependent RNA polymerase activities I and II: Dependence on food intake. Biochem. Biophys. Res. Comm. 54: 62-68. Bernstoin, P. and J. Ross (1989). Poly(A), poly(A) binding protein and the regulation of mRNA stability. Trends in Biochem. Sci. 14: 373-377. Bjerkvig, R., S.IC Steinsv~lgand O.D. La~rum (1986). Reaggregation of fetal rat brain cells in a stationary culture system. In Vitro Cell Dev. Biol. 22: 180-200. Bjtrkhem, I., U. Andersson, E. Sudjama-Sugiaman, G. Eggertsen and P. Hylemon (1993). Studies on the link between HMG-CoA reductase and cholesterol 7 alpha-hydroxylase in lymph-fistula rats: evidence for both transcriptional and post-transcriptional mechanisms for down-regulation of the two enzymes by bile acids. J. Lipid Res. 34: 149%1503. Blobel, G. and V. Potter (1966). Nuclei from rat liver: Isolation method that combines purity with high yield. Science. 154: 1662-1665. Brown, EL., E. Tahaoglu, G.J. Graham and JJ. Maio (1993). Inducible transcriptional activation of the human immunodeficieney virus long terminal repeat by protein kinase inhibitors. Mol. Cell. Biol. 13: 5245-5254. BSAC Working Party (1991). A guide to seusivity testing. J. Antimicrob. Chemother. 27: Suppl. D. Chang, C.D., L. Ottavio, S. Travali, K.E. Lipson and R. Baserga (1990). Transcriptional and posttrauscriptional regulation of the proliferating cell nuclear antigen gene. Mol. Cell. Biol. 10: 3289-3296. Chomczynski, P. and N. Saechi (1987). Single-step method of RNA isolation by acid guanidinium thiocyanate-phenol-chloroform extraction. Anal. Bioehem. 162: 156-159. Clarke, IJ., A. Rao, P.C. Fallest and M.A. Shupnik (1993). Transcription rate of the follicle stimulating hormone (FSH) beta subunit gene is reduced by inhibinin sheep but this does not fully explain the decrease in mRNA. Mol. Cell. Endocrinol. 91: 211-216. Coehary, E.F., Z. Kikinis and ICE. Paulson (1993). Positional and temporal regulation of lipogunie gene expression in mouse liver. C~meExpr. 3: 265-278. Crick, F.H.C. (1958). On protein synthesis. Biological replication of macromolecules. Symp. Soc. Exp. Biol. 12: 138-163. Davis, J.F. and M.R. Felder (1993). Mouse ethanol-inducible cytochmme P-450 (P450IIE1). Characterization of cDNA clones and testosterone induction in kidney tissue. J. Biol. Chem. 268: 16584-16589. Dehio, C. and J. Schell (1994). Identification of plant genetic loci involved in a pesttranscriptional mechanism for meioticaUy reversible transgene silencing. Prec. Natl. Acad. Sci. USA. 91:
20 5538-5542. Eader, L.A., L Gnsella, L. Dorman and H.A. Young (1994). Induction of multiple cytokine gene expression and IRF-1 mRNA by flavone acetic acid in a murine macrophage cell line. Ceil. Immunol. 157: 211-222. Eick, D., F. Kohlhuber, D.A. Wolf and LJ. Strobl (1994). Activation of pausing RNA polymerases in nuclear run-on-experiments. Anal. Biochem. 218: 347-351. Heming, A. (1929). On the antibacterial action of cultures of a penicillium, with special reference to their use in the isolation of B. Influenzee. Br. J. Exp. Pathol. 10: 226-236. Gad, S. (1993). Alternatives to in vivo studies in toxology. In: B. Ballantyne, T. Marrs and P. Turner, eds., General and Applied Toxicology, vol. 1, 179-206, New York, Stockton Press. Giere, R. (1991). Understanding Scientific Reasoning. Fort Worth, Harcourt Brace Jovanovich, Inc. Hacking, I. (1983). Representing and Intervening. Cambridge, Cambridge University Press. Hacking, I. (1992). The self-vindication of the laboratory sciences. In: A. Pickering, ed., Science as Practice and Culture, 29-64, Chicago, The University of Chicago Press. Hargrove, J.L and F.H. Schmidt (1989). The role of mRNA and protein stability in gene expression. FASEB J. 3: 2360-2370. Hirschelmann, R., C. v. Rein, A. Springer and U. Grife (1990). Mode of action of antiinflammatory agents: relation between arachidonic acid metabolism and hydrogen peroxide production. Eur. J. Pharmacol. 183: 2254-2255. Hirschelmann, R. (1991). Mode of action of antiinftammatory agents: The problem of data interpretation after experiments in vivo / ex vivo, in situ and in vitro. Naunyn-Schmiedeberg's Arch. Pharmacol. 344: Suppl. 2: Rl10. Hoekman, M.F., LM. Rientjes, J. Twisk, R.J. Planta, H.M. Princen and W.H. Mager (1993). Transcriptional regulation of the gene encoding cholesterol 7 alpha-hydroxylase in the rat. Gene. 130: 217-223. Hiittenrauch, R. and P. Speiser (1985). In vitro - in vivo correlation: An unrealistic problem. Pharmaceut~ Res. 3: 97-107. Jacob, S.T. and K.M. Rose (1977). RNA polymerases and poly(A)polymerase from neoplastic tissues and cells. Meth. Cancer Res. 14: 191-241. Kostuik, J.P. and T.J. Smith (1991). Pitfalls of biomechanical testing. Spine. 16: 1233-1235. Kubicki, A., K. Steinmuller and P. Westhoff (1994). Differential transcription of plastome-encoded genes in the mesophyll and bundle-sheath chloroplasts of the monocotyledonous NADP-malic enzyme-type C4 plants maize and Sorghum. Piant~ Mol. Biol. 25: 669-679. Lakates, I. (1970). Falsification and the methodology of scientific research programmes. In: I. Lakatos and A. Musgrave, eds., Criticism and the Growth of Knowledge, 91-196. London, Cambridge University Press. Latour, B. (1987). Science in Action. Milton Keynes, Open University Press. Lin, G., A.E. Pearson, R.W. Scamurra, Y. Zhou, M.J. Baarsch, D.L Weiss and M.P. Murtaugh (1994). Regulation of interleukin-8 expression in porcine alveolar macrophages by bacterial lipopolysaccharide. J. Biol. Chem. 269: 7%85. Lipton, P. (1991). Inference to the Best Explanation. London, Routledge. Marie, S., M.J. Diaz-Guerra, L Miquerol, A. Kalm and P.B. Iynedjian (1993). The pyruvate kinase gent as a model for studies of glucose-dependent regulation of gene expression in the endocrine pancreatic beta-cell type. J. Biol. Chem. 268: 23881-23890. Marzluff, W.F. and R.C. Huang (1984). Transcription of RNA in isolated nuclei. In: B.D. Haines and S.J. Higgins, eds. Transcription and Translation: A Practical Approach, 89-129. Oxford, IRL Press. Mill. LS. (1843). A System of Logic: Ratiocinative and inductive; being a connected view of the principles of evidence, and the methods of scientific investigation. London, John W. Parker and Son. National Research Council (1985). Models for Biomedical Research. A New Perspective. Washington DC, National Academy Press.
21 Paget, G.E. (1963). Experimental studies on the toxicity of Atromid with particular reference to the f'me structural changes in the liver of rodents. J. Atherosclerosis Res. 3: 729-736. Peirce, C.S. (1868). Some Consequences of Four Incapacities. J. Speculative Phil. 2; To be found in: C. Hartsthorne and P. Weiss, eds. (1960), Collected Papers of Charles Sanders Peirco, 5.264-5.314. Cambridge, Mass., Harvard University Press. Popper, K. (1980). The Logic of Scientific Discovery. 10th revised impression. London, Hutchinson. Ptashne, M. (1988). How eukaryotie transcriptional activators work. Nature 335: 683-689. Reid, L M . and D.M. Jefferson (3.984). Culturing hepatocytes and other differentiated cells. Hepatology. 4: 548-559. Rougvie, A.E. and J.T. Lis (1988). The RNA pelymerase II molecule at the 5" end of the uninduced hsp 70 gene of D. melanogaster is transcriptionally engaged. Cell. 54: 795-804. Salehi-Ashtiani, K. and E. Goldberg (1993). Differences in regulation of testis specific lactate dehydrogenase in rat and mouse occur at multiple levels. Mol. Reprod. Dev. 35: 1-7. Scala, R.A. (1987). Theoretical approaches to validation. In: A.M. Goldberg, ed., In Vitro Toxicology. Approaches to Validation, 1-9. New York, Mary Ann Liebort, Inc. Schilling, LJ. and P.J. Farnham (1994). Inappropriate transcription from the 5 ' end of the murine dihydrofolate reductase gene masks transcriptional regulation. Nucleic Acids Res. 22: 3061-3068. Skripal, I. and V.V. Babiehev (1989)o Increase of the in vitro activity of the mollicutes transcriptase using substances of organic nature. Mikrobiol Zhurnal, Kiev. 51: 19-23. Speth, C. and I. Oberbaumer (1993). Expression of basement membrane proteins: evidence for complex post-transcriptional control mechanisms. Exp. Cell Res. 204: 302-310. Stewart Gilmour, R. (1984). Transcription of chromatin. In: B.D. Haines and S.J. Higgins, eds., Transcription and Translation: A Practical Approach, 131-152. Oxford, IRL Press. Strand, R., R. Bee and T. Hatmark (1994). The choice of resuspension medium for isolated rat liver nuclei: Effects on nuclear morphology and in vitro transcription. Mol. Cell Bioehem. 139: 149157. Strobl, LJ. and D. Eick (1992). Hold back of RNA polymerase lI at the lranscription start site mediates down-regulation of c-myc in vivo. EMBO J. 11: 3307-3314. Swillena, S. and I. Pirson (1994). Highly sensitive control of transcriptional activity by factor heterodimedzation. Bioehem. J. 301: 9-12. Tam, S.P. and R.G. Deeley (1994). Regulation of apolipeptotein A-I gene expression by phenobarbital in the human hepatocarcinoma cell line, Hep3B. Atherosclerosis. 105: 235-243. Timmers, A.M., B.R. Newton and W.W. Hauswirth (1993). Synthesis and stability of retinal photoreceptor mRNAs are coordinately regulated during bovine fetal development. Exp. Eye Res. 56: 257-265. Tosic, M., A. Roach, J.C. de Rivaz, M. Dolivo and J.M. Matthicu (1990). Postttanscriptional events are responsible for low expression of myelin basic protein in myelin deficient mice: role of natural antisense RNA. EMBO J. 9: 401-406. Widnell, C.C. and J.R. Tam (1966). Studies on the stimulation by ammonium sulphate of the DNAdependent RNA polymerase of isolated rat liver nuclei. Biochim. Biophys. Acta 123: 478-492. Yount, G.L., P. Ponaalle and J.D. White (1994). Pentylenetetrazole-induced seizures stimulate transcription of early and late response genes. Brain Res. Mol. Brain Res. 21: 219-224. Zhou, H., S.S. Manji, D.M. Findlay, T.J. Martin, J.K. Heath and K.W. Ng (1994). Novel action of retinoic acid. Stabilization of newly synthesized alkaline phosphatase transcripts. J. Biol. Chem. 269: 22433-22439.
