Appearances Can Be Deceiving – The Problem With Genetic Knockout Mice Studies


When scientists want to study the effect that a gene has on an organism, they perform what is called a “knockout” experiment. The knockout method is an experimental method, usually employed in mice, whereby the expression of a specific gene under study is blocked. The goal of this method is to learn about what effect the absence of the gene has on the animal.

Due to the inevitable rise in the importance of genetic targeting in therapeutics, the knockout model has become a very important staple in studying the functions and endogenous expression patterns of single genes in vivo. While knockout experiments can produce valuable information about the gene of interest, as explained below, this is not always the case.  Thus, results from knockout experiments should be interpreted with caution.

There are three overall categories of problems, or “shortcomings”, with knockout studies. The first results from the limitations inherent in the knockout method technology and experimental method employed, the second results from the phenomena of biological robustness, and the third results from the complexity of the organism and constant interplay of a myriad of genes and proteins in vivo.

First Shortcoming

The first problem is that original genetic material from the embryonic stem cells used for the knockout study can remain in the genome, despite attempts to breed it out. These genetic confounding factors in turn impact the phenotype observed and thus the results obtained, since this residual genetic material can contribute to the observed phenotype; you don’t know whether the phenotype is due to single gene ablation or background genetic effects from residual genes.

There are three ways to improve this experimental shortcoming: 1) improve the gene targeting technology itself, 2) improve breeding techniques employed, or 3) rescue the knocked out gene during experimentation. Rescue is the most common method. If a gene is inactivated, and you reactivate it (i.e. “rescue” it), if the observed phenotype changes there’s a good chance it could be due to that gene.

According to Eisener-Dorman et al., for “any assessment of whether the ablation of a gene is, in fact, responsible for a phenotypic trait, the basic question should always be asked: is the observed phenotype relevant to what is known of the protein function? If the function of a gene is unknown, or if the observed phenotype deviates from what is reasonably anticipated, then the potential influence of [background stem cell] genes should be evaluated.”1

Second Shortcoming

Robustness is classified as a lack of variation in phenotype to genetic or environmental changes. As Dear et al. explain, “central metabolic pathways appear to have more alternatives than other pathways. This might reflect intrinsic robustness where central metabolic pathways must function under variable physiological and environmental conditions. However, it may also be adaptive—central metabolic pathways are critical for the organism’s survival and a back-up mechanism may be advantageous. It has been found in an analysis of transcriptional and signal transduction networks that parallel pathways connecting a regulator to a regulated molecule are not, as is commonly perceived, rare but are actually quite common…”.2

Genetic robustness can occur through “(i) genetic buffering—where alternative pathways for a process exist in the organism, or (ii) functional complementation—where genes are to some extent redundant in function. Two genes are considered to be redundant if they can fully or partially substitute each others functions.”

A common example of biological robustness is that often the same mutations that are studied between human and mouse in a specific gene result in diffrent phenotypes. That is, mouse mutations in gene X do not give the same result as human mutations in gene X (this is the primary problem in Pulmonary Hypertension… animal models, including knockout mice studies, don’t fully recapitulate the human form of PH): “Consider, for example, the OCRL1 gene, which encodes a phosphatidylinositol 4,5-bisphosphate 5-phosphatase. This gene is mutated in Lowe syndrome, a rare genetic disorder in humans that results in serious physical and mental problems. Yet, the mouse Ocrl knockout appears unaffected. However, mice have a related gene, Inpp5b, which is not present in humans. Inactivation of this gene results in only a mild phenotype, while the Ocrl-Inpp5b double knockout is embryonic lethal. Thus Inpp5b may be able to protect mice from any deleterious effects that would normally result from the absence of Ocrl.”

Finally, consider the example of the Hsp90 gene in Drosophila: “When this gene is mutated, widespread phenotypic variation results from other mutations, previously silent in the presence of the wild type Hsp90. Thus, Hsp90 is able to buffer against genetic mutations that would normally have a phenotypic effect.”

Third Shortcoming

The third and final shortcoming results from the complexity of the genome; there are a myriad of potential interactions between genes, between genes and the environment, and between proteins, and feedback loops exists between all of these…

In short, a gene might be doing something “X”, but that “X” combined with the product/activity of another gene “Y” produces an effect “Z”, which we will refer to here is a phenotype or microphenotype (disease, or a microstate like a malignancy or a lesion). Let’s consider two scenarios here…

Scenario 1: X is thought to be bad, Y is unknown, and Z is the bad resulting phenotype (a disease, etc.)… thus delete X and improve Z

In this scenario, Y is the unknown variable. Since Y is the factor you are unaware of, and you only see Z which is the combined effect of X and Y, you may think that Z is really caused ONLY by X. So, when you take away X, if Z disappears, you would be led to the conclusion that the gene you knocked out caused Z.

What’s the problem with this? You don’t really know that the gene product X is causing Z by itself; you don’t know it works in tandem with the product Y of another gene to produce the observed phenotype. Thus, in this case, therapies that target the gene that produces X will be unsuccessful because of the confounding factor Y that you are unaware of. What’s more is that these therapies may even have unintended side effects due to the fact that X may have other crucial functions critical to the organism’s survival and homeostasis.

Scenario 2: Loss of X is thought to be bad, Y is unknown, and Z is the bad phenotype… thus add/increase X to Improve Z

In this scenario, let’s use an example from PH. In PH, knockout of the BMPR2 gene can induce PH… and there are strong suggestions that loss of BMPR2 function is a principal factor in PH pathogenesis. To PH researchers credit, however, they do understand that to induce PH you usually need both BMPR2 loss of function AND an external stimulus like inflammation. However, just because there is an external stimulus doesn’t mean there isn’t another gene that is working in synergy with BMPR2 to prevent PH. That is, is BMPR2 really THE gene responsible for PH? Perhaps it is both the lack of BMPR2 and the presence of HERV-K genes? This would explain why inflammation can act as a coactivator in inducing PH, since HERV-K is a endogenous retrovirus responsible for releasing cytokines and inflammatory factors.

In this example, therapy for PH would include inducing BMPR2. But my point is, that even if you induce BMPR2, there could be another factor Y, silently working away to induce PH. This explains why people without BMPR2 mutations still can get PH. When you took out BMPR2 function, you saw the PH phenotype because you didn’t know about Y and that it was working to induce PH. Thus, an incorrect conclusion is that BMPR2 loss induces PH. However, even with BMPR2 working, Y is also working… and all you see is Z, which is PH.

In both of the above scenarios, Y can be any one of the myriad of human genes… This is why we should be wary of interpreting knockout mice studies. Analyzing one variable is important linear systems and simple systems… but in complex nonlinear systems like biology, it can lead to mistaken conclusions. As a result, results from knockout studies should be interpreted with caution, and researchers should fully understand all of the shortcomings associated with such experiments.

Hopefully, improvements in technology will improve the shortcomings of knockout mice studies. However, even with improvements in technology, scientists still face the daunting task of teasing out causal factors in experiments that involve manipulation of a complex environment such as an organism’s genome.

References & technical explanations of knockout methods:


Leave a Reply

Your email address will not be published. Required fields are marked *