text
stringlengths 10
379k
|
|---|
Ovarian cancer is the fourth leading cause of cancer-related deaths in women in the United States and the leading cause of gynecologic cancer deaths. The standard first-line chemotherapy for patients with ovarian cancer is a platinum-taxane combination regimen. In excess of 80 percent of patients will respond initially to this treatment. However, less than 10 percent will remain in remission [].
Recently, we described a sub-group of epithelial ovarian cancer (EOC) cells that express the protein Myeloid Differentiation Protein 88 (MyD88). MyD88 is an adaptor protein that is required for Toll-like receptor (TLR) signaling, a signaling pathway involved in inflammatory response to bacterial and viral products []. EOC cells that express MyD88 constitutively secrete pro-inflammatory cytokines and are resistant to the taxane paclitaxel, which is a known TLR-4 ligand. Upon TLR-4 ligation with either LPS or paclitaxel, these cells secrete higher levels of pro-inflammatory cytokines and, more importantly, proliferate in culture []. We hypothesized that MyD88 is a specific marker for paclitaxel resistance and is required for paclitaxel-induced cell growth in EOC cells. The objective of this study was to develop an optimized method that can detect MyD88 expression in ovarian cancer tumors and thus function as a biomarker for selection of therapy.
Proteomics has emerged as an important tool to study biological processes in both physiological and pathological circumstances. Identification of specific proteins as biomarkers for a disease or condition may be used for diagnosis and/or therapy. Currently, however, no optimized method exists that can sensitively detect ovarian cancer biomarkers from tumor tissue prior to initiation of therapy. A limiting factor in the discovery and analysis of potential marker from tumor samples are the “contaminating” signals originating from normal cells, including immune cells, that are infiltrating the tumor. In order to overcome this difficulty, several investigators have developed mechanical and/or chemical approaches to separate the cellular components of the tumor. However, the purity of these preparations has never reached the desired levels required for diagnostic.
The use of laser capture microdissection (LCM) has been proposed as an alternative method to obtain pure cell populations. This approach has been used successfully to identify RNA expression and DNA in stored samples [,]. However, determination of RNA in cells obtained by LCM is time consuming and requires special measures to prevent RNA degradation. In the present study, we describe a novel approach for the detection of protein expression in as low as 1,000 pure epithelial ovarian cancer cells dissected by LCM. Furthermore, we demonstrate that MyD88 expression in LCM-dissected ovarian cancer cells can be used as a marker for paclitaxel resistance.
Primary antibodies used for Western blot analysis were rabbit polyclonal anti-human and mouse antibody to MyD88 (dilution 1:1000 in PBS-T/1 percent FFPM) from eBioscience (San Diego, California), mouse monoclonal anti-human antibodies to FasL/CD95L (dilution 1:10,000 in PBS-T/1 percent FFPM) from BD Biosciences (San Jose, California), mouse monoclonal anti-human antibodies to hIAL/XIAP (dilution 1:10,000 in PBS-T/1 percent FFPM) from BD Biosciences PharMingen (San Diego, California). Secondary peroxidase-conjugated antibodies, anti-rabbit IgG (H+L) made in goat and anti-mouse IgG (H+L) made in horse, were purchased from Vector Laboratories (Burlingame, California).
In preparation for Laser capture microdissection, a 5x5 mm tissue sample was embedded in OCT at -20°C. Eight-micron sections were cut with the microtome and collected on Leica glass-foiled PEN-membrane slides. The slides were fixed in 95 percent ethanol for five minutes and stained with H&E for 30 seconds at RT, followed by washing in ddH₂O for 30 seconds. Dehydration is required for successful Laser microdissection, and this was accomplished by immersing the slides in 95 percent ethanol and then in Histosolve, a xylene substitute (Shandon, Inc., Pittsburgh, Pennsylvania), 15 minutes each. The slides were air dried at RT for 15 minutes. Following dehydration, to prevent protein degradation, the samples were either used immediately for microdissection or they were stored in air-tight plastic wrapping at -20°C.
Eight-micron sections were cut with the microtome, collected on Leica glass-foiled PEN-membrane slides and fixed in 95 percent ethanol for five minutes as described above. The slides were then placed in a staining dish with 0.1 percent H₂O₂ in 0.1M PB for 10 minutes to quench for endogenous peroxidase activity and then washed in wash buffer three times. Wash buffer was prepared by adding 0.01 percent Triton X-100 (Sigma Aldrich, St. Louis, Missouri) to 0.1M PB. The area around the tissue sections was scored with a Pap pen to limit the amount of antibodies and reagents used. All steps occurred at room temperature with the slides placed in a moisture chamber to keep the tissue from drying out during the procedure. To block for non-specific background, 100-200 µl of 3 percent BSA made in wash buffer were added to the circumscribed areas and incubated for 20 minutes in the moisture chamber. Next, the primary antibody was incubated for two hours followed by washing with wash buffer (each wash is two to three minutes times three), the specimens were incubated with Biotin labeled secondary antibody for 30 minutes. After this incubation, the slides were washed once with wash buffer then twice with 0.1M Tris (pH 7.5). The tissue was incubated with Streptavidin-HRP conjugate for 30 minutes and then washed with 0.1 M Tris three times. The color was developed with DAB (2.5 mg DAB in 5 ml 0.1 M Tris). Twenty five µl of 0.03 percent H₂O₂ were added to the chromogen just before use. At the end of the procedure, the slides were washed with double distilled water and counterstained with Mayer’s hematoxylin (Sigma Aldrich, St. Louis, Missouri). The specimens were dehydrated and then stored as described above.
Proteins were resolved under reducing conditions on 12 percent SDS-PAGE gels and then transferred onto polyvinylidene difluoride paper (NEN Life Sciences, Boston, Massachusetts) []. Membranes were blocked at room temperature for one hour with 5 percent FFPM in PBS/0.05 percent Tween 20 (PBS-T). After three washes for 10 minutes, each with PBS-T, membranes were incubated overnight at 4°C with the primary antibody in PBS-T/1 percent FFPM. After incubation with the primary antibody, membranes were washed three times with PBS-T and then incubated at room temperature with the appropriate secondary antibody conjugated to peroxidase (Vector Laboratories) in PBS-T/1 percent FFPM. After three washes for 10 minutes each with PBS-T and three washes for 10 minutes each with distilled water, the peroxidase-conjugated antibody was detected by enhanced chemoluminiscence (PerkinElmer Life Sciences, Boston, Massachusetts). The signal intensity was analyzed using a Kodak digital imaging analysis system (Kodak Image Station 4000MM) and Kodak Molecular Imaging Software (Scientific Imaging Kodak Company).
Our first objective was to determine whether we could detect protein expression by Western blot analysis in microdissected cells obtained from ovarian tumor samples that were snap-frozen in liquid N2. Thus, we microdissected increasing number of cells with a maximum of 8,000 cells and a minimum of 100 cells from 8 µm ovarian cancer sections and then analyzed them for the expression of MyD88 and other cancer–related proteins such as XIAP [-] and FasL [,] []. As shown in , a strong signal for MyD88 expression was observed in samples containing 8,000 and 5,000 microdissected cells, and no signal was detected for samples containing less than 1,000 cells.
In order to determine whether the sensitivity of the assay remains similar for different proteins, we evaluated the expression of XIAP and FasL using the same number of cells. Contrary to MyD88, XIAP and FasL expression is detected in samples containing as little as 1000 and 100 cells, respectively [].
Tumors are heterogeneous tissues containing normal supporting cells, immune cells and cancer cells. Immune cells infiltrate the stroma and may “contaminate” the protein profile of the tumor sample []. Indeed, when we analyzed MyD88 expression in H&E stained samples without using appropriate markers for cell type identification, we found that MyD88 was expressed in all samples. The explanation for these results is that the immune cells present in the tumor tissue and surrounding the cancer cells express MyD88 []. In an H&E stained sample, the selection of specific cells based on their morphology alone is not sufficient to identify intra-tumoral infiltrated cells. An advantage of the LCM is that it affords the possibility of selecting a specific group of cells based on the expression of cellular markers. Therefore, our next objective was to determine whether immunohistochemistry staining affects the detection of proteins in microdissected samples. Thus, we compared the sensitivity of protein detection of cells obtained from H&E stained samples with cells obtained from immuno-stained samples. Tumor tissues were immuno-stained with either antibodies for cytokeratin 7 (epithelial marker) or CD45 (pan-leucocyte marker). Specific staining for cytokeratin 7 and CD45 was observed in these frozen sections [ and data not shown]. When we compared MyD88 and XIAP expression in cells obtained from these two preparations, no difference was observed in the signal intensity for these proteins between the two groups, suggesting that the process of IHC does not affect the samples’ integrity []. Once we established that IHC does not affect protein detection, we evaluated MyD88 expression in ovarian cancer samples after staining for CD45 (panleukocyte antigen). As shown in , when we dissected only the cancer cells (CD45 negative cells) and avoided the immune cells (CD45 positive cells), MyD88 was differentially expressed in tumors from different patients with identical pathological diagnoses.
We evaluated 20 tumor samples from patients with epithelial ovarian cancer. The patient characteristics are summarized in . Eleven out of 20 tumors were MyD88 positive and nine tumors were MyD88 negative. Figures and show two representative samples after CD45 staining in preparation for the laser capture microdissection and MyD88 expression in pure cancer cells.
A statistical significant correlation between assay prediction of response and progression-free interval (PFI) was observed in all these cases []. Furthermore, as shown by the Kaplan-Meier curve in , the patients that had MyD88 negative tumors responded better to treatment and had a significantly longer survival when compared with the patients that had MyD88 positive tumors []
This study challenges the current approach of prescribing chemotherapy by relying solely on the pathological diagnosis. Instead, we propose that determination of molecular markers be undertaken and incorporated into the diagnostic process prior to treating patients with neoplastic diseases of the ovary. We show that the status of MyD88 expression may have a significant impact in chemo-responsiveness and therefore may be an indispensable piece of information to obtain prior to commencing treatment.
Although the link between inflammation, NF-κB, and cancer progression has been well characterized, the specific signaling pathway upstream of NF-κB is currently not well elucidated. As mentioned above, MyD88 is one of the proteins required in TLR4-mediated early phase NF-κB activation []. Our previous studies [] showed that Type I epithelial ovarian cancer (EOC) cells constitutively secrete IL-6, IL-8, GROα, and MCP-1. In response to LPS stimulation of TLR4, these cells proliferated and showed marked increase in the secretion of these cytokines via the activation of NF-κB. In contrast, type II EOC cells do not constitutively secrete cytokines, and no changes in cell proliferation or cytokine production were observed following treatment with LPS. The characteristic pro-inflammatory environment generated by Type I EOC cells was lost upon the knockdown of MyD88, suggesting that an active MyD88-dependent TLR4 signaling is responsible for the LPS-induced, NF-κB-mediated EOC cell proliferation and cytokine secretion.
As mentioned above, paclitaxel, a first-line chemotherapeutic agent used in the treatment of ovarian cancer [-], is a known TLR4 ligand [,,]. Thus, we also determined if paclitaxel treatment would induce the same response in type I EOC cells as was seen with LPS. Our results showed that similar to LPS, paclitaxel also induced NF-κB activation, increased secretion of pro-inflammatory cytokines, and proliferation of type I EOC cells. In contrast, no cytokine secretion was observed in type II EOC cells, and more importantly, they underwent apoptosis in response to paclitaxel. These results suggest that an activated MyD88-dependent TLR4 signaling pathway to NF-κB confers EOC cells the ability to promote a pro-inflammatory environment and the development of paclitaxel resistance. It is noticeable of the specificity of the signaling pathway to paclitaxel, since treatment of type I EOC cells with carboplatin, a chemotherapeutic drug which is not a TLR4 ligand, did induce neither proliferation nor cytokine production []. The mechanism that regulates MyD88 expression in these two types of cells, and the correlation of MyD88 expression and clinical response to paclitaxel in patients with EOC, however, still remains to be determined.
Moreover, endogenous ligands of TLR4 such as HMGB1, heparan sulfate and polysaccharide fragments of hyaluronan, which are released from the damaged tissue and necrotic cells, may also contribute to cancer progression and paclitaxel resistance [-].
Tumors contain different cell populations that are incorporated in a heterogeneous structure. These different cell types can be studied after meticulous isolation. LCM represents an alternative approach to obtaining pure cancer cells from fresh-frozen cancer tissue, thus providing an accurate snapshot of the tumor and its microenvironment . After lysis, these cells can be analyzed and their protein expression characterized by Western blot, ELISA, and Luminex assays.
Early identification of chemoresistance in patients with ovarian cancer has important treatment implications. It allows tailoring of therapy to individual patients and prevents the use of cytotoxic drugs with no therapeutic benefit while avoiding toxicities. Unfortunately, no available method exists to predict the chemoresponse of cancer cells prior to initiation of therapy.
The standard treatment for ovarian cancer is the combination of carboplatin and paclitaxel. Unfortunately, approximately 15 percent of newly diagnosed patients will not respond to this combination, and their disease will progress during or shortly after completion of chemotherapy. This percentage of non-responders increases to 65 to 75 percent for recurrent cancers.
Based on these studies, we hypothesized that MyD88 expression can be used as a marker for paclitaxel-resistance. However, we found that a limiting factor in the detection of MyD88 in whole tumors was the presence of intra-tumoral immune infiltrate. Immune cells, mainly macrophages, express high levels of MyD88 and may give a false positive result.
Utilizing LCM and immunostaining protocols that allow clear identification of the desired cell types can overcome this obstacle. In our experiments, CD45 staining proved to be highly useful for the identification and selection of pure cancer cells from tumor tissue and classification of tumors as either MyD88 positive or negative.
We found that proteins expressed in high-levels, such as FasL or XIAP, can be detected with as low as 1,000 cells. However, for proteins with lower expression levels, like MyD88, a minimum of 5,000 cells was required for detection.
The advent of the LCM, used in conjunction with protein or gene detection techniques, has enabled us to gain a better perception of the complex interactions that occur within and among cells in living tissues. In this study, we have found that fresh-frozen tissue is a suitable material for this procedure. We have demonstrated in this report how proteins involved in the apoptotic cascade and chemoresistance of epithelial ovarian cancer cells can be accurately detected using a three-step approach. First, tumor specimens are snap-frozen in liquid nitrogen and eight µm thick specimens are stained to facilitate the selection of pure tumor cells. Second, target cells are identified based on their immunostaining and collected with the LCM system. The third step involves lysis of the collected cells followed by Western blot or Luminex multiplex assay.
The protocol presented here provides a fast and easy method for analyzing protein expression in tissues after LCM. We have used this method to evaluate the correlation between MyD88 expression and clinical outcome, and we have found that all patients who had MyD88 positive tumors presented with poor progression-free interval and overall survival after chemotherapy with carboplatin and paclitaxel, while the patients with MyD88 negative tumors had an excellent response to chemotherapy. A prospective clinical study to validate the specificity and sensitivity of this test aimed at identifying paclitaxel-resistant patients is under way. We have recently developed an immunohistochemistry protocol for MyD88 that is faster, less labor intensive, and can be applied to paraffin sections [].
In summary, we describe a sensitive method to determine MyD88 as a potential marker for chemoresponse to paclitaxel. We have identified a specific subclass of cancer patients who, based on their MyD88 expression, are paclitaxel resistant. The use of molecular markers like MyD88 will enable us to design an individualized of treatment, thereby eliminating toxicity from agents without therapeutic benefit or even potential harm, as evidenced by the stimulating effect of paclitaxel on MyD88 positive EOC. |
Of the 4,000 cases of acute lymphoblastic leukemias (ALL) diagnosed in the United States every year, one third occur in adults. Adult ALL remains a difficult disease to cure in contrast with the cure rate of almost 80 percent in children and adolescents []. Front-line protocols often call for aggressive therapy for high-risk patients during induction, consolidation, and maintenance therapy. Allogeneic hematopoietic stem cell transplantation is offered to eligible patients with the promise of a prolonged leukemia free survival []. For patients who relapse after transplantation or who are not eligible for stem cell transplant due to lack of an appropriate donor or comorbid medical conditions, novel salvage therapies are desperately needed []. Recently, two nucleoside analogs, nelarabine and clofarabine, have been FDA approved for patients with relapsed or refractory T-cell and B-cell lymphocytic leukemia respectively [-].
Clofarabine is a novel purine nucleoside analog that has shown promise in early phase II studies for pediatric patients with B-ALL. The product of rationally designed improvements to two older adenine nucleosides, clofarabine is taken up into cells by an active nucleoside transporter [] and is subsequently phosphorylated into nucleotide analogues clofarabine 5’-mono-, di-, and triphosphate, with clofarabine triphosphate being the active metabolite []. Clofarabine has been shown to have three mechanisms of action in leukemia cells: first, it is incorporated into DNA and impairs DNA elongation and repair; second, it is a potent inhibitor of ribonucleotide reductase, depleting the nucleotide pool primarily of dCTP and dATP; and third, it may directly induce apoptosis by altering the mitochondrial membrane and subsequently enabling release of cytochrome c [].
In a phase II study of clofarabine in a pediatric population with refractory B-ALL, the response rate was 30 percent []. In the adult population, an initial phase I study showed that two of 13 patients responded []. In a subsequent phase II study in patients with refractory or relapsing ALL, acute myelogenous leukemia (AML), and myelodysplastic syndrome (MDS), two of 12 patients with ALL had responses []. In this report, we describe a case wherein low-dose weekly administration of clofarabine induced a response in a patient with refractory T cell ALL with primarily cutaneous disease.
JK is a 35-year-old male with precursor T-lymphoblastic leukemia diagnosed in 2001 with a bone marrow blast immunophenotype of CD2+CD7+CD3+ cells with aberrant expression of CD13 and negative staining for TdT and myeloperoxidase. His karyotypic analysis was indicative of a 4;11 translocation involving the MLL gene at 11q23. He was induced and underwent allogeneic stem cell transplantation (SCT) from his HLA-identical fraternal twin brother in April 2002. He was in remission with limited skin chronic graft-vs.-host disease until January 2003, when he relapsed and was treated with ICE chemotherapy followed by a donor lymphocyte infusion. He remained in remission until February 2005, when he developed nodular violacious skin lesions which on biopsy revealed T-cells consistent with his original leukemia. Full restaging including bone marrow biopsy and cerebrospinal fluid analysis were negative, indicating an isolated skin recurrence. Re-induction therapy was initiated with cyclophosphamide and gemcitabine but was complicated by significant myelosuppression and renal insufficiency. He received a brief course of systemic retinoid therapy with bexarotene and spot electron beam irradiation for symptomatic improvement of nodular disease on his face with rapid recurrence at the completion of radiation. He then was treated with nelarabine in February 2006 and responded but developed worsening renal insufficiency.
He quickly developed recurrent nodular tumors over his brow, around his right ear, on his chest, and on his back and chest []. On restaging in October 2006, the patient had no evidence of lymphadenopathy or involvement of liver or spleen by CT scan. A bone marrow biopsy revealed relapse of known precursor-T lymphoblastic leukemia involving 40 percent of the marrow. The cells were CD2+ CD7+ CD34+ HLA-DR+ CD3-. Flow cytometry of the blood was negative for blasts with only occasional reactive lymphocytes. At that time, his leukocyte count was 2.0 with 64 percent neutrophils and 35 percent lymphocytes, the hematocrit was 24.5 and platelets were 52 x 10³/dl. He initiated therapy with clofarabine at 10 mg/m2 weekly for three consecutive weeks every 28 days. Treatment schedule and toxicities are shown in Table 1. There was no hepatic toxicity, and only mild fluctuations of the creatinine were noted. Treatment was held on the third week of the first two cycles due to neutropenia.
After two cycles of clofarabine, the patient’s lesions regressed significantly as shown in . Cycle 3 was delayed due to a hospitalization for herpetic esophagitis, which resolved with acyclovir therapy. He remains on weekly low-dose clofarabine with ongoing response after three months. His platelet count has been stable at 99,000/mm³. An unrelated donor has been identified, and the patient will undergo a second allogeneic stem cell transplant.
In this case report, we demonstrate the efficacy and safety of weekly low dose clofarabine in a patient with relapsed and refractory precursor T-ALL. Because clofarabine is renally excreted, it has been dosed with caution in patients with renal or hepatic insufficiency []. In this case, clofarabine was successfully administered, albeit at a reduced dose, without worsening of renal function or evidence of disproportionate drug-related toxicity.
The observed response in this patient is surprising, given the low dose and intermittent scheduling []. Pharmacokinetics studies showed that intracellular concentrations did not become saturated at doses less than 20 mg/m²; furthermore, despite an estimated half-life of 24 hours, there was no evidence of accumulation in doses less than 40 mg/m²/day. studies demonstrated that breakthrough DNA synthesis occurs in leukocytes derived from patients receiving systemic doses less than 40 mg/m², suggesting suboptimal effect []. Consistent with the predictions made by the data and the pharmacokinetics of the drug, the clinical trials have suggested better results with higher doses. In particular, patients with relapsed CLL were treated with 2 mg/m² without any responses observed (ILEX Products Inc., personal communication). Nevertheless, it may be reasonable in patients with more rapidly proliferating disorders to administer clofarabine at a lower dose using this weekly schedule.
Lastly, we present this case report as the first evidence for the efficacy of clofarabine in treating primary or secondary cutaneous lymphoma/leukemia. As mentioned above, this patient had extensive involvement of the skin with his T-cell leukemia. The cutaneous response was brisk with this low dose of clofarabine. One possible explanation for the sensitivity of the patient’s tumor to low-dose clofarabine is localization of the drug in the skin, as was demonstrated in murine pharmacokinetic models by Lindemalm et al. Skin rash and palmoplantar erythrodysesthesias have been reported in the clinical trials in which clofarabine is administered on a five-day schedule at higher doses, perhaps related to retention of the drug in skin. Based on this experience, it may be worthwhile to further investigate the efficacy of low dose clofarabine in primary and secondary cutaneous malignancies. |
MicroRNAs are small, non-coding RNAs of 18 to 24 nucleotides, discovered in 1993 in the nematode . Lee et al. [] found that a gene crucial for post-embryonic development, , does not code for a protein, but rather is transcribed into a 22-nucleotide RNA molecule. This molecule could repress the expression of the mRNA by directly interacting with its 3' untranslated region (UTR) []. While this was recognized as a new method of gene regulation, it was initially considered as an oddity peculiar to . The discovery of , another small RNA involved in developmental timing in [], and the finding that both its sequence and temporal expression pattern were largely conserved in a variety of organisms [] immediately suggested these small RNAs might in fact play important and conserved roles in gene regulation, and the identification of hundreds of miRNAs in the worm, fly, and mammalian genomes followed [].
To date, more than 300 miRNAs have been discovered in humans, and computational analyses predict that up to 1,000 miRNAs exist in the genome [,]. Since miRNAs can regulate more than one target, estimates indicate they may be able to regulate up to 30 percent of the protein-coding genes in the human genome [], highlighting their importance as global regulators of gene expression.
MiRNAs are transcribed by RNA Polymerase II into large precursor RNAs, often several kilobases in length, called pri-miRNAs. Notably, the majority of human miRNAs are transcribed from regions found within introns of either protein-coding or non-protein-coding transcripts, but a minority is found in isolated regions of the genome, within the exons of noncoding mRNA genes, or within the 3'UTRs of mRNA genes.
In the nucleus, these pri-miRNAs are capped and polyadenylated prior to being processed by Drosha, a member of the RNase III enzyme family, in conjunction with the double-stranded RNA-binding protein DGCR8/Pasha. This processing step produces segments ∽70 nucleotides in length, which fold into stem-loop structures known as pre-miRNAs []. These are exported from the nucleus in a GTP-dependent fashion by exportin 5 and are subject to an additional processing step by another RNase III enzyme, Dicer. This step releases a double-stranded RNA duplex, ∽22 nucleotides in length, which in turn is incorporated into the miRISC complex, in analogous fashion to that observed in RNA interference (RNAi). In this complex, the mature miRNA strand is retained, and the complex is now capable of regulating its target genes.
The identification of miRNA target transcripts remains one of the greatest challenges in the field today. MiRNAs can be organized into families based on sequence homology, which is particularly strong at the 5' end of the mature miRNA, suggesting this section of the mature transcript has been preserved through evolution and thus plays an important role in the process of target recognition. Indeed, studies have shown this 5' region, often called the "miRNA seed," to be crucial for both the stability of the mature miRNA and its incorporation into the miRISC complex [-]. Bioinformatic approaches have taken advantage of this "miRNA seed" to predict miRNA targets across the genome. It has been predicted that a single miRNA can bind over 200 different target transcripts, and, notably, these targets are highly diverse, from transcription factors to transporters [-].
There is nevertheless evidence supporting a role for the regions not encompassed within the "miRNA seed." Perhaps the clearest example of this is found in , whose complete mature miRNA sequence is conserved across species [], suggesting an important role for the 3' region of the miRNA in target transcript binding. Further insights into the miRNA-target recognition mechanism are needed, as they will make target prediction much more accurate and efficient.
Cancers of all types share a number of characteristics, such as the loss of cellular identity, an increased ability to grow and proliferate, and alterations in the systems controlling cell death. Studies performed in a variety of organisms have revealed that miRNAs have the ability to regulate these cellular processes, suggesting that they could be involved in cancer. For instance, and control the timing of developmental events in . Mutations in these miRNAs resulted in abnormalities in cell-cycle exit as well as in the execution of a terminal differentiation program, preventing cells from reaching their fully differentiated state [].
In , it has been found that over-expression of the Bantam miRNA causes excessive growth of wing and eye tissue by blocking the pro-apoptotic action of the gene [,]. It was also found that another miRNA, miR-14, can act as a strong suppressor of apoptosis. Deletion of this miRNA caused increased expression of the apoptotic effector caspase, , suggesting a direct regulation of this factor by miR-14 []. Likewise, the miR-2/6/11/13/308 family of miRNAs has been reported to induce widespread apoptosis in embryos through the regulation of the proapoptotic factors , , , and , which act through the inhibition the caspase inhibitor Diap1 in response to a number of natural and toxic conditions [].
Another aspect in which miRNAs appear to have a role is the proper differentiation of cells into different tissues. For example, miR-273 and the miRNA encoded by have been shown to play important roles in the patterning of the nervous system [,], and miR-430 has been implicated in brain development []. The importance of miRNA in development has been shown also in mammalian systems, examples of which are miR-181 in the differentiation of hematopoietic cells toward the B-cell lineage [], miR-374 in pancreatic islet-cell development [], miR-143 in adipocyte differentiation [], miR-196 in limb patterning by SHH [], and miR-1 in heart development [].
The first study directly suggesting that dysregulation of miRNA played an important role for miRNAs in tumorigenesis came from Calin et al. [] It was known that loss of chromosomal region 13q14 was strongly associated with B cell chronic lymphocytic leukemia (CLL), the most common form of leukemia in the Western hemisphere; however, no gene or genes had been definitely directly linked with CLL. After refining the critical region within 13q14 to about 30kb, Calin et al. failed to identify any protein coding genes but recognized that two miRNAs, miR-15a and miR-16-1, were clustered within this chromosomal locus. Indeed, expression of these miRNAs was found to be diminished or completely ablated in > 65 percent of CLL cases examined. Further studies revealed a germline C-to-T mutation seven base pairs downstream of the miR-16-1 hairpin in two out of 75 CLL patients. This mutation was not found in any of 160 control individuals. Because the mechanisms for miRNA biogenesis are not yet fully understood, it is unclear what the precise effect of this mutation is; however, it was found that the mutation correlated with diminished expression of this miRNA in patient-derived CLL cells, while preventing the processing of the miRNA into its mature form when expressed in heterologous cell systems [].
The common loss of miR-15a and miR-16-1 in CLL, as well as the loss of 13q14 in mantle cell lymphoma (50 percent of cases), multiple myeloma (16 to 40 percent) and prostate cancer (60 percent) [,], strongly suggests that these two miRNAs act as tumor suppressor genes. While their full target complement is unknown, they appear to mediate their effects largely by down-regulating the anti-apoptotic protein BCL2. This protein is often found expressed at high levels in CLL and is thought to be important for the survival of the malignant cells []. Consistent with a role for these miRNAs in the down-regulation of BCL2, loss of miR-15a and miR-16-1 correlates with elevated levels of BCL2, while heterologous expression of these miRNAs results in decreased levels of the endogenous proteins []. The 3'UTR of the transcript has been found to contain binding sites for miR-15a and miR-16-1, and reporter constructs containing the 3'UTR are downregulated upon co-expression with miR-15a and miR-16-1. Moreover, expression of these miRNAs is capable of inducing apoptosis in a leukemia cell line. Further research is necessary to better evaluate whether miR-15a and miR-16-1 exert their effects mainly through BCL2 or target additional transcripts in CLL. Nonetheless, the evidence available supports an important role for miR-15a and miR-16-1 in the prevention of lymphomagenesis and leukaemogenesis.
As noted earlier, the miRNAs of the family were among the first to be described, and experiments performed in provided the first indication that they may be involved in cancer. Early studies showed that loss-of-function mutants were defective in the transition from late larval to adult stage. Upregulation of in adult seam cells, on the other hand, is necessary to induce cell-cycle exit and terminal differentiation. If this miRNA is absent, these cells fail to differentiate, undergoing additional rounds of division, a phenomenon often observed in cancer cells [].
There are 12 homologs in the human genome, organized in eight clusters. These clusters map to fragile sites associated with lung, breast, urothelial, and cervical cancers, suggesting that family members may act as tumor suppressors. At least four of these clusters have been confirmed to be commonly lost in malignancies []. Studies by Takamizawa et al. and Yanaihara et al. have presented evidence that transcripts of certain homologs are significantly downregulated in human lung cancer and that low levels of correlate with poor prognosis [,]. Moreover, it has been shown that transient expression of miRNAs of the family in cell lines derived from lung adenocarcinomas inhibited colony formation, suggesting that these miRNAs could have growth-suppressing properties and opening the possibility that might be used as a therapeutic tool.
Important insights into the mechanism through which is capable of controlling cellular proliferation came from our laboratory, derived from the observation that the family member miR-84 negatively regulates , the ortholog of human , , and in []. We have shown that certain members of the family are capable of genetically interacting with RAS, and and RAS are reciprocally expressed in lung tumor samples. We also found that is capable of negatively regulating RAS in human cells, and overexpression of in human cancer cell lines results in reduced levels of RAS protein, compared to untreated cells. As expected, if levels are reduced in cancer cell lines, levels of RAS protein increase significantly. Moreover, through the use of reporter constructs, it was confirmed that the regulation of RAS by takes place through the direct interaction of the miRNA with the 3'UTR of Ras mRNA.
The human RAS family of proteins is one of the most important components of a myriad of signaling cascades, has been shown to possess oncogenic activity, and is often mutated in tumors []. As such, these studies position as a promising therapeutic tool for the treatment of lung cancer, as well as other malignancies resulting from the overexpression of RAS. It will be important to more carefully analyze the specific contributions of the individual members of this miRNA family, in order to better understand the -mediated regulation of cell growth.
Studies have implicated other miRNAs in tumorigenesis. MiR-143 and miR-145, for instance, have been shown to be constantly down-regulated in colorectal tumors [], and recent studies by Croce et al. also have shown that the downregulation of these miRNAs is a common occurrence in breast carcinomas and breast cancer lines []. Moreover, the location of these miRNAs is known to be in a genomic locus frequently deleted in myelodysplastic syndrome []. While some studies have ascribed the down-regulation of miR-143 and miR-145 in colon cancer to a block in Dicer processing [], further studies are required to determine whether the expression levels of these miRNAs are specifically altered.
p
The multiple lines of evidence indicating that microRNAs are differentially expressed in normal and in tumor samples suggest that miRNAs could serve as useful tumor profiling tools. Indeed, profiling studies have shown that each cancer type possesses a distinct miRNA expression signature, and this signature can provide useful information about the malignancy. The adaptation of high-throughput technologies have facilitated the study of the expression of multiple miRNAs in a given sample, making it possible to profile substantial sample numbers with relative ease. Among the most important methods for miRNA profiling is the use of oligonucleotide miRNA microarray analysis, which allows the researcher to simultaneously determine the expression levels of hundreds of miRNAs in a given sample [-]. Multiple variations of this technique have been developed, underscoring its position as the central miRNA profiling tool [-]. Other important approaches to the problem are also available, such as quantitative PCR for precursor miRNAs [] or mature miRNAs [,]; genome-wide approaches with serial analysis of gene expression (SAGE), such as miRAGE []; and bead-based flow cytrometric techniques [].
Taking advantage of these tools, several groups have studied the miRNA expression of several types of cancer. Michael et al. [] and Cummins et al. [] have studied the expression profile of the known miRNAs in colorectal cancer samples. These studies independently found a general down-regulation of miRNA levels in tumor cells compared to normal colonic epithelium, while also observing a significant decrease in the levels of miR-143 and miR-145 beyond this general effect. Cummins et al. found a signature of 50 differentially expressed miRNAs in malignant vs. normal colonic epithelium, of which 32 miRNAs had reduced levels in tumor cells. Yanaihara et al. studied over 100 patient-matched pairs of primary malignant and normal adjacent lung tissue and found that the expression of 43 miRNAs was significantly different in the tumor tissues compared to the normal adjacent tissues. Of these miRNAs, 28 were down-regulated and 15 were up-regulated in the malignant tissue []. Similar studies have been carried out in CLL [], breast cancer [], glioblastoma [], pancreatic cancer [], hepatocellular carcinoma [], and thyroid papillary carcinoma [].
The ability to profile the miRNA expression of a tumor with accuracy and reproducibility in a clinical setting would constitute a valuable medical tool. Indeed, miRNA profiles have been shown to be highly informative. Using a bead-based approach, Lu et al. attempted to classify 17 poorly differentiated tumors with non-diagnostic histological appearance based on their miRNA expression profiles. A clinical diagnosis on these tumors was established by anatomical context, whether directly (a primary tumor observed in a given organ) or indirectly (a metastasis of a previously identified primary tumor). They were able to correctly diagnose 12 of the 17 samples, against 1/17 correct diagnoses for an analysis using their mRNA profile []. While the number of transcripts analyzed was much smaller for miRNAs than for mRNAs (~200 against > 15,000), it is possible that the miRNA-based diagnoses are more accurate because of the regulatory role of these molecules, while most mRNAs will not be regulatory in nature. Further studies, however, are necessary to ascertain the value of miRNAs as diagnostic tools, and their potential value as prognostic markers, by correlating miRNA expression with type-specific parameters such as metastatic potential, proliferative index, and response to existing treatments.
While the field of miRNAs — and particularly the study of the roles of miRNAs in cancer — is at an early stage, their potential as targeted therapeutic tools has not gone unnoticed. A promising approach is to target oncogenic miRNAs with oligonucleotides complementary for either their mature or precursor sequence, deemed anti-miRNA oligonucleotides (AMOs). A number of chemical modifications for these oligonucleotides are available, with the purpose of making them more stable and less toxic. 2'-O-methylated AMOs are perhaps the most common form of modified RNA, conferring limited nuclease resistance, and enhancing the binding affinity to RNA. Inhibition of function with 31-nucleotide 2'-O-Me AMOs has been demonstrated in HeLa cells and larvae [], and 2'-O-Me AMOs have been reported to abrogate miRNA function in cellular assays [,], providing proof-of-principle on the blockage of miRNA function by extraneous oligomers. Another common modification to the structure of AMOs is that of the locked nucleic acid (LNA)-modified oligomers. In these RNAs, the 2' oxygen atom is linked to the 4' position through a methylene bridge, forming a rigid bicyclic structure locked into a C3'-endo (RNA) sugar conformation []. This modification results in greatly enhanced affinity for RNA and the formation of exceedingly strong duplexes with excellent mismatch discrimination properties [-]. This makes them prime candidates for AMO-based therapy, and results in rodents have been encouraging [].
While the use of AMOs could be useful to inhibit oncogenic miRNAs, another therapeutic possibility concerns the cases in which certain miRNAs are under-expressed, such as in lung cancer, or miR-15a and mir-16-1 in CLL. In this case, synthetic miRNAs — likely modified as discussed above — in either their mature or precursor form could be delivered to the target cells to counteract the deficiency and halt malignant growth. While this approach has been attempted in cell culture [], further testing in animal models is required.
A pressing challenge to the use of miRNAs as therapeutic tools concerns finding methods to deliver synthetic miRNAs or AMOs to the desired tissues in a targeted and effective manner. Given that these molecules cannot discriminate between healthy and malignant cells, side effects of treatment remain a concern. Likewise, since these studies are in their infancy, little is known about the pharmacokinetics of AMOs or synthetic miRNAs, limiting the discussion of their promise as therapeutic tools into the speculative arena. Upon reaching a deeper understanding of the mechanism of miRNA biogenesis and action, and the development of new delivery technologies, these small RNAs might well fulfill their promise as valuable therapeutics. |
The breakthrough that allowed the sequencing of viral genomes (and later, larger cloned genomes) was the use of restriction endonucleases to manipulate and analyze DNA. Bacterial cells synthesize restriction enzymes to defend themselves against invading foreign DNA, and these enzymes were first identified and characterized on the basis of their ability to cleave bacteriophage DNA.
The value of restriction enzymes was established by a series of pioneering studies by Daniel Nathans at Johns Hopkins and Paul Berg at Stanford on the genome of SV40, a small virus that normally infects monkeys. SV40 was discovered in tissue cultures of monkey kidney cells in which poliovirus vaccine was produced, and many vaccinated individuals were inadvertently inoculated with SV40 during early poliovirus vaccination programs.
Enhanced interest in SV40 lay in the fact that it and the closely related mouse polyomavirus caused tumors in rodents. It was hoped that detailed analysis of tumor viruses would shed light on the genesis of common human cancers and suggest new approaches to prevent and treat cancer.
This hope has been realized many times over. SV40 has been implicated in some human cancers, including childhood brain tumors, mesotheliomas, and non-Hodgkins lymphoma, but an etiologic role of SV40 in these tumors has not been established unequivocally []. Nevertheless, the early development of molecular biology techniques to analyze SV40 DNA led to the first restriction map and the first localization of a genetic signal, the viral origin of DNA replication, in eukaryotic cells [-].
The first recombinant DNA molecule constructed consisted of SV40 linked to bacteriophage lambda DNA [].
This work set the stage for much of modern genetics and molecular biology and won Nobel prizes for Nathans and Berg. Studies on SV40 also played a major role in the genetic engineering revolution and the birth of the biotechnology industry.
Prior to the use of restriction endonucleases, sequencing studies were largely limited to the handful of bases at the very ends of viral genomes (a striking limitation in the case of circular genomes). Now the entire viral genome was accessible.
The initial large-scale sequencing efforts bore fruit in 1977 with the publication of the complete DNA sequence of ɸX174 by a consortium headed by Frederick Sanger (who already had won a Nobel Prize for developing methods to sequence proteins and later won a second Nobel for DNA sequencing) [].
The next year, the DNA sequence of the tumor virus SV40 was solved by groups headed by Sherman Weissman at Yale University and Walter Fiers in Belgium [-].
ɸX174 and SV40 are relatively simple viruses, each with only about 5,000 DNA base pairs (compared to the 6 billion base pairs in each of our cells), but with the primitive methods available at that time, this sequence analysis required many years and prodigious effort. With today’s technology, the genomes of these small viruses can be knocked off in a busy afternoon.
With the support of the “War on Cancer” declared by the National Cancer Institute in the 1970s, tumor virologists rapidly adopted sequencing as a primary method to analyze viral genomes, and larger DNA tumor viruses soon fell to sequencing efforts: papillomaviruses, the cause of warts and cervical and some other human cancer, with 8,000 base pairs, in 1982; and adenoviruses, which cause respiratory tract disease in humans and tumors in experimental animals, with 36,000 base pairs, in 1984. The following year, 1985, saw the completion of the 172,000 base pair sequence of the DNA of Epstein-Barr virus, a herpesvirus that infects virtually all humans and causes certain infected individuals to develop Hodgkin’s disease, Burkitt’s lymphoma, or nasopharyngeal carcinoma. Since the sequencing was done manually, this heroic feat required eight postdoctoral scientists working full time for several years. The poxviruses, a virus group that includes the dreaded smallpox virus and related viruses that can cause tumors in animals, weighed in at 191,636 base pairs in 1990.
The necessity of sequencing and analyzing ever-larger viral genomes spurred improvements in technology. The first sequences were derived from RNA copied from the viral DNA, but researchers soon moved to sequencing the DNA itself. A key breakthrough was the development of rapid enzymatic DNA sequencing methods by Sanger and colleagues at Cambridge University [] replacing more cumbersome chemical methods invented by Maxam and Gilbert at Harvard University []. The Cambridge group also developed “shot-gun” methods for sequencing the genome of Epstein-Barr virus, in which the sequence of random fragments of viral DNA was determined, and the sequences of overlapping DNA segments were then aligned by computer. This method turned out to be essential for solving the sequences of the much larger bacterial and cellular genomes, and eventually the postdoctoral fellows were replaced by robots for the cloning operations and by automated DNA sequencing machines. Our ability to record and analyze long DNA sequences was also sorely tested, a predicament that helped lead to the development of methods of computational analysis required for the study of bacterial and cellular genomes.
The sequences of viral genomes revealed the great promise and limitations of large-scale sequence analysis. For the first time, the entire genetic legacy of an organism was manifest. Genes and regulatory signals were identified, protein structure and function was deduced, and overall genetic organization was laid bare. It seemed possible that a life form could be reduced to a set of simple instructions, thus revealing its most intimate secrets.
Novel human viruses were discovered based on DNA sequencing; genes that viruses captured from cellular DNA were identified; and pathogenic mechanisms and the genetic and evolutionary relationships between different viruses were elucidated. Again, studies of tumor viruses led the way.
Kaposi Sarcoma herpesvirus, which commonly causes cancer in AIDS victims, was discovered solely on the presence of novel viral DNA sequences, related to Epstein-Barr virus in a tumor [], and hepatitis C virus was defined based on sequence information [,]. Hepatitis C virus and its compatriot hepatitis B virus are responsible for the development of hepatocellular carcinoma, the third most common cancer in the world [,]. Analysis of oncogenic retroviruses isolated from animal tumors revealed that these viruses carried genes responsible for cancer and that these genes were actually altered versions of cellular oncogenes [], a discovery that won the Nobel Prize for Michael Bishop and Harold Varmus at the University of California at San Franscisco. Extensive sequencing also resulted in the identification of multiple different types of human papillomaviruses and the discovery that certain high-risk HPV types cause important human cancers [,].
These studies ultimately led to development of prophylactic vaccines designed to prevent virus infection and the resulting human cancers. Vaccines based on the major capsid protein of the high-risk human papillomaviruses will prevent much cervical cancer, and the hepatitis B virus surface antigen vaccine already is causing a reduction in the incidence of hepatocellular cancer [,]. These cancer vaccines are produced by using recombinant DNA technology with defined viral DNA segments because of the difficulty of propagating these human tumor viruses in the laboratory. These two vaccines alone have the potential to eliminate greater than 10 percent of the world-wide cancer burden. The prevention of cancers by vaccination is the culmination of a century-long effort to tie infectious agents to cancer and validates this entire enterprise []. In addition, ongoing work at Yale and elsewhere demonstrates that viral gene expression is required to maintain the survival and growth of certain cancers, suggesting that anti-viral drugs may be useful in treating these cancers once they develop [,,]. Vaccination against viral non-structural antigens expressed by cancer cells also has emerged as a potential therapeutic approach.
Viruses responsible for acute infections have been subjected to intense sequencing efforts. Complete sequence information enabled the laboratory synthesis of infectious poliovirus, a small virus that has so far eluded sustained efforts for eradication []. In a remarkable example of molecular archaeology, sequencing of archived samples of the 1918 pandemic influenza virus, which caused the most deadly acute epidemic in history, allowed the resurrection of this virus []. Studies based on this work will have a profound effect on our understanding of these important human pathogens and the genetic basis for future influenza pandemics, but this work also inspired much discussion about the potential misuse of sequence information and scientific knowledge. The very rapid identification of a novel coronavirus as the causative agent of SARS was a stunning achievement based entirely on sequencing [,]. Recent sequencing studies also revealed that the environment harbors a vast number of uncharacterized bacteriophage and other viruses [,]. Remarkably, the genes contained in these viruses often do not resemble previously known genes, hinting at immense genetic diversity of viruses.
It was anticipated that genomic sequence information would lead to the rapid elucidation of all the details of virus reproduction and myriad interactions between viruses and cells and suggest new rational approaches to combat virus infections and virally induced tumors. But there also were complications. Viruses were more clever, versatile, and enigmatic than we appreciated, and not all their secrets were revealed by simple sequence analysis alone. RNA splicing, first discovered in oncogenic adenoviruses and widespread in cellular genes as well [,], results in the synthesis of proteins assembled from segments encoded by DNA scattered about the genome, complicating efforts to deduce protein sequence from DNA sequence.
In other cases, viruses selectively ignored signals that should tell the cellular machinery where genes started and stopped, and RNA editing or ribosomal frameshifting confounded the definition of a gene (e.g., [,]). One of the first viral sequences solved revealed the startling situation that the same piece of DNA could encode two proteins with entirely different amino acid sequences []. Thus, it became clear that complete understanding of an organism, even one as simple as a virus, cannot be deduced from sequence information alone, but the services of experimental biologists were still needed.
The great value of sequencing was immediately clear to the scientific community and inspired successful attempts to decipher larger cellular genomes.
The first complete DNA sequence of a bacterial genome, that of , was reported in 1995 [], and now hundreds of bacterial genomes have been decoded, providing valuable information about the lifestyle, evolution, and pathogenesis of these organisms. And in 2001, the entire human genome sequence was completed.
In parallel, the genomes of several other pathogens and organisms useful as experimental models, including the malaria parasite, insects, worms, plants, and other mammals, have been sequenced. These sequences uncovered a treasure trove of information that is revolutionizing our understanding of evolution, gene organization and function, cellular function, development, and disease. These studies also revealed the disquieting fact that a substantial fraction of our own genomes is derived from remnants of viruses related to oncogenic retroviruses. It has even been postulated that the transfer of viral genes into primordial RNA-based cells gave rise to the three branches of cellular life on earth: bacteria, archea, and eukaryotes [].
DNA sequencing has had a particularly profound effect on our understanding of carcinogenesis and on efforts to control this disease. Sequencing of the cellular versions of the oncogenes first identified in tumor viruses revealed newly arising mutations in many sporadically occurring human cancers [], helping to establish the notion that much of cancer has a genetic basis.
Many of these oncogenes encode growth factors, growth factor receptors, signal transduction proteins, and transcription factors, which are crucial components of signaling pathways that determine how cells respond to their environment, grow, and die. Based on these and related studies, it is now possible in some cases to predict a patient’s prognosis or response to chemotherapy based on the sequence of cellular proteins involved in cell signaling and growth [].
The elucidation of the human genome sequence also allowed the construction of microarrays that have been used to assess the transcriptional profile of normal and cancer cells. These studies revealed the existence of subsets of various tumors with markedly different prognosis, suggesting underlying biological differences that can be exploited in developing improved approaches for managing these diseases [].
In addition, advances in rapid sequencing and computational analysis allowed comprehensive comparison of the sequences of normal and cancer cell genomes and the identification of crucial shared mutations in independently arising cancers []. The genes identified in this manner are likely to play important roles in carcinogenesis and provide new targets for therapy and diagnosis.
Sequencing of DNA isolated from cancers also revealed the frequent occurrence of mutations in genes encoding tumor suppressor proteins, particularly p53, which was first identified as a binding partner of SV40 large T antigen, the viral DNA replication protein [,]. Large T antigen also binds the retinoblastoma tumor suppressor protein, a central component of another tumor suppressor pathway often inactivated by mutation in human cancers []. Strikingly, three unrelated groups of DNA tumor viruses, SV40, human papillomaviruses, and adenoviruses, all inactivate these same tumor suppressor proteins, implying the existence of a limited number of regulatory nodes essential for viral replication and cellular carcinogenesis.
These important cellular genomics efforts clearly had their origins in studies of viral genomes. Francis Collins, who heads the National Human Genome Research Institute and directed the United States’ effort to sequence the human genome, was trained in the Weissman laboratory at Yale that sequenced SV40. Two of the prime movers of the landmark 1995 sequence were Hamilton Smith and Craig Venter. Smith and his Johns Hopkins’ colleague Nathans supplied essential information and materials for the SV40 sequencing project, and a year before the appearance of the first complete bacterial sequence, Venter published the sequence of smallpox virus DNA.
The ability to decode bacterial and cellular genomes is a watershed achievement in biology, one that is transforming our concepts of microbial pathogenesis, cellular function, evolution, and human disease, particularly cancer. This remarkable accomplishment is the logical extension of pioneering studies begun a quarter of a century earlier on viruses, the simplest organisms at the border between chemistry and life. |
To study the structure of the high and low MW CSP forms, we made polyclonal antisera to peptides representing the entire NH-terminal and COOH-terminal thirds of CSP from , a rodent malaria parasite. These antisera recognized the appropriate full-length peptides (Fig. S1 A, available at ) and did not recognize peptides representing the central repeat domain (Fig. S1 B). In addition, the NH-terminal antiserum did not recognize the COOH-terminal peptide and the COOH-terminal antiserum did not recognize the NH-terminal peptide (unpublished data). Western blot analysis of a sporozoite lysate showed that the NH-terminal antiserum recognized only the high MW CSP form, indicating that all or part of the NH terminus is proteolytically cleaved to generate the low MW CSP form ( B). In contrast, mAb 3D11 (which recognizes the repeat region) and the COOH-terminal antiserum recognized both CSP forms.
To determine what class of protease was responsible for cleavage, we performed pulse-chase metabolic labeling experiments in the presence of different protease inhibitors. We labeled sporozoites with [S]Cys/Met and chased with cold medium containing the indicated inhibitor ( C). In the absence of protease inhibitors, ∼80% of labeled CSP was cleaved after 2 h. In the presence of the metalloprotease inhibitor 1,10 phenanthroline or the aspartyl-protease inhibitor pepstatin, there was no effect on CSP processing. In addition, EDTA had no effect on CSP processing, indicating that divalent cations are not required. -transepoxysuccinyl-leucylamide-[4-guanido]-butane (E-64), a highly specific cysteine protease inhibitor, and PMSF, a serine protease inhibitor, inhibited CSP processing. Leupeptin and TLCK, inhibitors of both cysteine and serine proteases, also inhibited processing. Although PMSF has been reported to have inhibitory activity against some papain family cysteine proteases (), it is a prototypical serine protease inhibitor. To further examine the role of serine proteases, we assayed two other serine protease inhibitors, aprotinin and 3,4 dichloroisocoumarin (3,4 DCI). Aprotinin inhibits most classes of serine proteases and would be predicted to inhibit the serine proteases of , which are subtilisin-like (). 3,4 DCI is a serine protease inhibitor that has some activity against cysteine proteases but does not react with papain-like cysteine proteases (). Neither compound had an effect on CSP processing.
We also performed pulse-chase metabolic labeling experiments with the human malaria parasite, , and found that E-64 inhibited CSP processing in this species ( D). These data suggest that CSP cleavage occurs by a similar mechanism in both rodent and human species.
To ensure that the protease inhibitors were not toxic to sporozoites, we incubated sporozoites with the different inhibitors and added propidium iodide, a dye that is excluded by viable cells but penetrates membranes of dying cells. The percentage of sporozoites that took up the dye in the presence of any of the protease inhibitors was no different from controls (unpublished data). In addition, we tested whether sporozoites incubated with protease inhibitors were less metabolically active. Analysis of CSP synthesis after sporozoites had been incubated with individual inhibitors for 2 h showed that it was not affected by E-64, leupeptin, or PMSF (Fig. S2, available at ).
Our data suggest that the processing enzyme is a cysteine protease. The cysteine proteases found in parasites are members of two clans, CA (papain-like) and CD (legumin-like) (for review see reference ), which can be distinguished by their sensitivity to E-64. The protease that cleaves CSP is inhibited by E-64 and, therefore, is a Clan CA, papain family cysteine protease. However, we found that PMSF, a serine protease inhibitor, also inhibited processing. As stated before, PMSF has been reported to have activity against papain family cysteine proteases and this could explain its inhibitory activity in our processing assay. Nonetheless, it is also possible that CSP cleavage is a complex multistep process involving distinct proteases.
To determine where CSP is cleaved, we mapped the epitopes recognized by the NH-terminal antiserum using overlapping peptides. As shown in A, the NH-terminal antiserum recognized peptides interspersed throughout the NH-terminal third of the protein, suggesting that the processed form lacks this entire region. These data raised the intriguing possibility that region I, found at the end of the NH terminus, contained the cleavage site. To test this, we used a recombinant parasite in which the last 21 amino acids of the NH terminus and the entire repeat region had been replaced by the orthologous region from CSP [Pf/Pb sporozoites; B and reference ). A Western blot of Pf/Pb sporozoites shows that both CSP forms are present, suggesting it is processed ( C). We performed pulse-chase metabolic labeling experiments with Pf/Pb sporozoites and found that after a 4-h chase, 50–80% of the high MW CSP is processed to the low MW form (unpublished data). When we tested whether E-64 could inhibit processing of the hybrid CSP, we found that it did ( D), indicating that the same protease cleaves both the native and hybrid CS proteins.
These data suggest that the cleavage site is found within region I because this sequence remains unchanged in the hybrid protein. Although it is possible that the cleavage site is outside of the swapped region, this is unlikely because the NH-terminal antiserum, which recognized peptides throughout the NH-terminal third of CSP, did not recognize the low MW CSP form. Previous studies have shown that the difference in size, by SDS-PAGE, between the high and low MW forms is ∼8–10 kD (, , –). The NH-terminal portion of CSP, beginning after the signal sequence and ending just before the repeat region, is predicted to be this size.
We investigated the cellular location of CSP processing. Immunofluorescence experiments with live sporozoites showed that they were recognized by the NH-terminal antiserum, demonstrating that full-length CSP was on the surface ( A). To confirm this, we biotinylated sporozoites expressing GFP with a reagent that does not enter cells. As shown in B, the high MW CSP form is biotinylated, indicating that it is on the surface. As a control, we immunoprecipitated GFP, an intracellular protein, and found that it was not labeled ( C). These findings are in agreement with a previous paper that showed that high MW CSP was on the surface of sporozoites () and suggest that processing occurs on the sporozoite surface.
In contrast with our findings, other investigators found that the majority of CSP on the surface was the low MW form, and concluded that processing occurred intracellularly (, ). In these studies, CSP was immunoprecipitated from sporozoites that were metabolically labeled and trypsinized. When compared with controls, trypsin-treated sporozoites were primarily missing the low MW CS band, indicating that the high MW CSP form was intracellular. However, in these experiments, trypsin was added immediately after labeling, which may not have allowed sufficient time for export of all the labeled CSP to the sporozoite surface. To investigate whether this was the case, we repeated this experiment and incorporated a chase into the experimental design. Sporozoites were metabolically labeled and kept on ice or chased in the presence of cyclohexamide to prevent further protein synthesis. Next, they were treated with pronase or pronase plus an inhibitor cocktail. As shown in D, if the parasites were kept on ice after labeling, the high MW CSP was not digested by pronase. However, if sporozoites were chased before pronase treatment, both CSP forms were digested, indicating that both forms were found on the sporozoite's surface, making this the likely location of processing.
Sporozoites isolated from salivary glands of infected mosquitoes are invariably contaminated with mosquito debris, raising the possibility that the protease that cleaves CSP is of mosquito origin. To address this question, we dissected and purified sporozoites in the presence of E-64, and then metabolically labeled them in medium without E-64. Cysteine proteases of mosquito origin would be extracellular and, therefore, irreversibly inhibited by the E-64 present during sporozoite isolation. However, we found that CSP was processed with the same kinetics regardless of whether sporozoites were purified in the presence or absence of E-64. These data suggest that the protease was synthesized (or secreted) after the removal of E-64 and, therefore, was of sporozoite origin ( E).
Proteolytic cleavage of cell surface and secreted proteins occurs during invasion of erythrocytes by the merozoite stage of (for review see reference ). To determine whether CSP cleavage was required for sporozoite entry into cells, a variety of protease inhibitors were tested for their ability to inhibit sporozoite invasion of a hepatocyte cell line. As shown in A, E-64 inhibited invasion by 90% and PMSF and leupeptin also had inhibitory activity. Pepstatin had no effect on invasion and the serine protease inhibitors aprotinin and DCI, which do not have activity against the papain family cysteine proteases, also did not have inhibitory activity on invasion. Importantly, pretreatment of target cells with E-64 had no inhibitory effect on sporozoite invasion. The ability of E-64 to inhibit invasion was not restricted to sporozoites, as invasion by both and sporozoites was also inhibited by E-64. Notably, the number of extracellular sporozoites was always enhanced in the presence of E-64, suggesting that there was an accumulation of attached sporozoites that were prevented from entering ( B). Because attachment to cells is a distinct stage of sporozoite invasion (), these results suggest that E-64 specifically blocks invasion and that attachment to cells does not require proteolytic cleavage of CSP.
These data suggest that CSP is cleaved during cell invasion. Therefore, we predicted that intracellular sporozoites would lose their reactivity to the NH-terminal antiserum, which recognizes only full-length CSP. However, we found that the majority of sporozoites associated with cells lost their reactivity to the NH-terminal antiserum regardless of whether they were intracellular or extracellular (unpublished data). In the absence of cells, 80–90% of sporozoites stained with this antiserum (unpublished data), suggesting that cell contact was the trigger for CSP cleavage. To test this, sporozoites were preincubated with cytochalasin D (CD), an inhibitor of sporozoite invasion but not attachment to cells (), in the presence or absence of E-64 and added to cells. As shown in , sporozoites incubated with CD plus E-64 stained with the NH-terminal antiserum, whereas those incubated with CD alone did not. mAb 3D11, directed against the repeat region of CSP, bound to both E-64–treated and untreated sporozoites. Controls in which sporozoites were incubated without cells showed that neither elevated temperature nor serum alone had a significant effect on CSP cleavage ().
In this assay, sporozoites were incubated with cells for only 2 min before being fixed and stained. The rapid loss of reactivity to the NH-terminal antiserum indicates that there is a dramatic increase in the kinetics of CSP cleavage when parasites are added to cells. In the absence of cells, the half life of newly synthesized CSP is ∼1 h ( and references , ). These data indicate that the secretion of the protease that cleaves CSP is regulated. It is likely that the low level cleavage observed in the absence of cells is due to leaky secretion from apical organelles, whereas exocytosis of larger amounts of protease is mediated by specific signals that are transduced upon contact with target cells.
It has been shown that sporozoites interact with cells in two distinct ways: they either rupture the plasma membrane and migrate through a cell or they enter with a vacuole and productively invade the cell (). To study whether CSP processing was preferentially associated with one of these processes, we tested whether E-64 inhibited sporozoite migration through cells. Migration can be quantified by including a high MW fluorescent tracer in the medium because it will enter cells that are wounded by sporozoites as they pass through. As shown in C, E-64 had no effect on sporozoite migration through cells.
These data indicate that CSP cleavage is associated with productive invasion of cells and suggests that sporozoites differentially recognize cells that they will invade; a finding that makes sense given that, in vivo, they travel through several cell barriers to reach their target, the hepatocyte. One question raised by these findings is how do sporozoites recognize hepatocytes? Previous work has shown that CSP binds to heparan sulfate proteoglycans (HSPGs) found on hepatocytes, making these molecules likely candidates for target cell recognition (for review see reference ). We are currently investigating whether binding of CSP to HSPGs triggers cleavage and initiates the cascade of events leading to productive invasion of cells.
Lastly, we tested E-64 as an inhibitor of malaria infection in vivo using a rodent model of the disease. Using a quantitative PCR assay, we compared the amounts of parasite rRNA in the livers of mice pretreated with E-64 or buffer and infected with sporozoites. We found that mice injected with E-64 were completely protected from malaria infection ( D). Although inhibitors of cysteine and serine proteases have not yet been used for the treatment of human disease, animal studies have shown the feasibility of using these inhibitors as drugs in the treatment of parasitic infections (for review see references , ). Our finding that we can completely prevent malaria infection by targeting the cysteine proteases of the sporozoite stage could lead to the development of new prophylactic agents for malaria.
These data are part of a growing body of work demonstrating that proteolytic processing of secreted and surface proteins is required for cell invasion by and other Apicomplexan parasites such as (, , ). One of the most well-studied examples is MSP-1, the major surface protein of merozoites, the infective form of the erythrocytic stage (for review see reference ). Interestingly, both CSP and MSP-1 have known cell-adhesive domains in their COOH termini, raising the possibility that cleavage controls the exposure of these domains. In CSP, the COOH terminus contains the TSR, a known cell-adhesive sequence that has been shown to bind with high affinity to HSPGs (for review see reference ). Previous studies have shown that the NH-terminal portion of CSP also binds to HSPGs (). Our data suggest a model for CSP cleavage that explains why this protein has two heparin-binding domains. Our hypothesis is that an initial interaction between cell surface HSPGs and the NH-terminal portion of CSP cross-links the protein and provides the signal for cleavage. In turn, cleavage exposes the cell-adhesive TSR, which binds with high affinity to HSPGs, initiating a cascade of events that ultimately lead to cell entry.
mAb 3D11 is directed against the repeat region of CSP (); mAb NYS1 is directed against the repeat region of CSP (); and mAb 2A10 is directed against the repeat region of CSP (). For immunoprecipitations, mAbs 3D11 and 2A10 were conjugated to sepharose as outlined previously (). Antisera to the NH- and COOH-terminal thirds of CSP were generated in rabbits using peptides that were provided by G. Corradin and M. Roggero (Institute of Biochemistry, Lausanne, Switzerland). The sequences of the NH- and COOH-terminal peptides were GYGQNKSIQAQRNLNELCYNEGNDNKLYHVLNSKNGKIYIRNTVNRLLADAPEGKKNEKKNKIERNNKLK and NDDSYIPSAEKILEFVKQIRDSITEEWSQCNVTCGSGIRVRKRKGSNKKAEDLTLEDIDTEICKMDKCS, respectively. Overlapping peptides and repeat peptides were synthesized and purified by Midwest Bio-Tech.
, –expressing GFP (), and recombinant sporozoites expressing a hybrid – CSP (Pf/Pb sporozoites; reference ) were grown in mosquitoes. –infected mosquitoes were obtained from D. Carucci (Naval Medical Research Center Malaria Program, Silver Spring, MD). Where indicated, sporozoites were purified by passage through two 3-μm polycarbonate membranes (Whatman).
Peptides were coated onto wells of Immunlon 2HB microtiter plates (ThermoLabsystems) and blocked, and antisera were added at the indicated dilutions. Binding was revealed with anti–mouse or anti–rabbit Ig-conjugated to alkaline phosphatase followed by the fluorescent substrate, 4-methylumbelliferyl phosphate and fluorescence was read in a Fluoroskan II plate reader.
or where indicated, or Pf/Pb sporozoites, were metabolically labeled in DMEM without Cys/Met, 1% BSA, and 400 μCi/ml L-[S]Cys/Met for 1 h at 28°C and chased in DMEM with Cys/Met and 1% BSA at 28°C in the presence or absence of the indicated protease inhibitor. For the pronase experiment, sporozoites were metabolically labeled in medium without BSA for 45 min at 28°C, washed, and resuspended in DMEM with Cys/Met and 100 μg/ml cycloheximide for 10 min and kept on ice or chased at 28°C for 1 h. Sporozoites were resuspended in 100 μg/ml pronase, ± pronase inhibitor cocktail (500 μg/ml antipain, 30 μg/ml aprotinin, 600 μg/ml chymostatin, 5 mg/ml EDTA, 5 μg/ml leupeptin, 10 mg/ml AEBSF, 7 μg/ml pepstatin, and 2 mM PMSF; reference ) for 1 h at 4°C, washed, and lysed in lysis buffer with pronase inhibitor cocktail and 1% BSA; CSP was immunoprecipitated.
Metabolically labeled sporozoites were lysed in lysis buffer (1% Triton X-100, 150 mM NaCl, 50 mM Tris-HCl, pH 8.0) with protease inhibitors for 1 h at 4°C, and lysates were incubated with mAb 3D11 agarose overnight at 4°C and washed with lysis buffer and lysis buffer with 500 mM NaCl and preelution buffer (0.5% Triton X-100, 10 mM Tris-HCl, pH 6.8). CSP was eluted with 1% SDS in 0.1 M glycine, pH 1.8, neutralized with Tris-HCl, pH 8.8, and run on a 7.5% SDS–polyacrylamide gel under nonreducing conditions. For experiments with or Pf/Pb sporozoites, a 10% SDS–polyacrylamide gel was used. Gels were fixed, enhanced with Amplify (Amersham Biosciences), dried, and exposed to film.
Sporozoite lysates were separated by SDS-PAGE, transferred to PVDF membrane, and incubated with either 4 μg/ml mAb 3D11, NH-terminal antiserum (1:3,000), COOH-terminal antiserum (1:3,000), or 4 μg/ml mAb 2A10 followed by anti–mouse or anti–rabbit Ig conjugated to horseradish peroxidase (HRP; 1:100,000). Bound antibodies were visualized using the enhanced chemiluminescence detection system (ECL).
transgenic for GFP was biotinylated using sulfo-succinimidyl-6′-(biotinamido) hexanoate according to the manufacturer's instructions (Pierce Chemical Co.). Lysates of biotinylated sporozoites were immunoprecipitated with either mAb 3D11 or polyclonal antibodies to GFP (1:200; Molecular Probes) followed by protein A coupled to agarose beads, loaded onto a 4–12% Tris-Glycine gel, transferred to PVDF, and incubated with either mAb 3D11 followed by anti–mouse Ig HRP, anti-GFP Ig (1:500) followed by anti–rabbit Ig HRP, or streptavidin–HRP (1:100,000). Bound antibodies were visualized using ECL.
Live sporozoites were incubated with NH-terminal antiserum (1:500 in DMEM/BSA) at 4°C for 2 h, washed at 4°C, and allowed to air dry on slides at 4°C. They were incubated with anti–rabbit Ig-FITC, washed, and mounted.
Invasion assays were performed as described previously (), with some modifications. For assays with and , Hepa 1–6 cells (CRL-1830; American Type Culture Collection) were used, and for assays with , HepG2 cells (HB-8065; American Type Culture Collection) were used. Sporozoites were preincubated with the indicated protease inhibitor for 2 h at 28°C and plated on cells in the continued presence of the inhibitor for 1 h at 37°C. In a control, Hepa 1–6 cells were incubated with 10 μM E-64 for 2 h at 37°C, the medium was removed, and untreated sporozoites were added. After incubation with sporozoites, cells were washed and fixed, and sporozoites were stained with a double-staining assay that distinguishes between extracellular and intracellular sporozoites.
sporozoites were incubated in DMEM ± 10 μM E-64 at 4°C for 2 h and added to Hepa 1–6 cells on glass coverslips. 30 min before sporozoites were added to coverslips, CD was added to all samples (final concentration, 1 μM). Sporozoites were centrifuged onto coverslips (1,250 ) for 5 min at 4°C. Coverslips were brought to 37°C for 2 min, fixed with 4% paraformaldehyde, and stained with either mAb 3D11 followed by anti–mouse Ig FITC or the NH-terminal antiserum followed by anti–rabbit Ig FITC. When sporozoites expressing GFP were used, the cells were only stained with the NH-terminal antiserum. As a control, sporozoites were spun onto coverslips without cells using the aforementioned protocol.
Sporozoites were preincubated ±10 μM E-64 for 2 h at 28°C and added to Hepa 1–6 cells in the continued presence of inhibitor with 1 mg/ml rhodamine-dextran. After 1 h at 37°C, the cells were washed and fixed, and rhodamine-positive cells were counted as outlined previously ().
Swiss/Webster mice were given three i.p. injections of DMEM ± E-64 (50 mg/kg/injection) at 16, 2.5, and 1 h before i.v. injection of 15,000 sporozoites. 40 h later, livers were harvested, total RNA was isolated, and malaria infection was quantified using reverse transcription followed by real-time PCR using primers that recognize –specific sequences within the 18S rRNA as outlined previously (). 10-fold dilutions of a plasmid construct containing the 18S rRNA gene were used to create a standard curve.
Fig. S1 shows the specificity of the NH- and COOH-terminal antisera as determined by ELISA. Fig. S2 shows that the protease inhibitors that inhibited CSP processing are not toxic to sporozoites. Online supplemental material is available at . |
To visualize the detailed microanatomy of brain immunological synapses, we set up an experiment in which T cells would selectively target virally infected brain astrocytes. To do so, nonreplicating adenoviral vectors were chosen to infect brain cells within a restricted site within the rat brain (i.e., the striatum) and infected cells were identified through expression of a reporter gene encoded within the vector (e.g., HSV1-thymidine kinase [TK]). More than 85% of infected cells were GFAP-expressing astrocytes (). 30 d later, animals were immunized with a systemic injection of RAd-HPRT (), an adenovirus encoding an unrelated transgene to that expressed in the brain, to stimulate a specific systemic immune response against adenovirus. The systemic administration of adenovirus is needed because the initial delivery of a nonreplicating adenovirus to the brain parenchyma fails to prime a systemic immune response against adenovirus (or other infectious antigens), as a result of the so-called immune privilege of the brain (–). A replication-deficient adenovirus was used to avoid any interference from viral replication on cell viability. In this model, cell loss is exclusively immune mediated, rather than as a result of viral replication.
Systemic immunization against adenovirus causes a specific infiltration of the brain injection site with T cells, and clearance of TK-expressing astrocytes and adenoviral genomes from the striatum ( and ). Astrocytes displayed MHC-I on their plasma membrane, thus constituting a potential target for activated CD8 T cells () (). Selective antibody depletion of CD4 and CD8 T cells 3 d after the systemic immunization demonstrated that both were necessary for astrocyte clearance (). That each type of T cell makes a separate contribution to clearance was shown by the fact that CD4 T cells remained confined to the perivascular compartment (), whereas CD8 T cells did infiltrate the site within the brain parenchyma proper where infected astrocytes were located (). CD8 T cells also established frequent close anatomical contacts with infected brain cells (). CD8 T cells' influx into the central nervous system (CNS) reached its peak at 14 d after immunization, whereas the loss of infected cells occurred between 14 and 30 d ( and ). The analysis of brain immunological synapses in vivo is based on the detailed study of brains from 25 rats and a total of at least 60 immunological synapses that were thoroughly studied in detail.
Polarization of phosphorylated tyrosine kinases, the initial stages in the formation of immunological synapses in vivo, can be detected preceding and throughout the clearance of infected astrocytes by CD8 T cells. To further characterize immune–target cell interactions and the activation of T cells, we studied the distribution of phosphorylated Lck and ZAP-70 in CD8 T cells contacting target astrocytes. Engagement of the TCR on CD8 T cells by MHC-I peptide complexes present on the surface of target APCs activates effector T cells and stimulates the T cell tyrosine kinase signaling cascade (, , ). Tyrosine kinase signaling activation was assessed by immunocytochemistry using antibodies recognizing specifically active, phosphorylated-Lck (p-Lck) (recognizing pY394), or active, phosphorylated-ZAP-70 (p-ZAP-70) (recognizing pY319) (, , , ).
Phosphorylated-Lck and p-ZAP-70 were polarized in CD8 T cells to sites of close membrane apposition between CD8 T cells and target astrocytes only in the virally injected hemisphere ( and ). Phosphorylation of tyrosine kinases precedes the establishment of mature immunological synapses. Thus, we studied the anatomical arrangements of these early sites of T cell–astrocyte interactions through serial reconstruction appositions identified with the confocal microscope, and three-dimensional artistic rendering of the confocal images ( and ). Most CD8 T cells displayed a single site of close membrane apposition with astrocytes (), although in some cases a CD8 T cell would make up to three such polarized contacts with a single target astrocyte (). CD8 T cells formed complex close interactions with target astrocytes. T cells either surrounded the target cell's processes ( [b, d, and e]) or displayed cytoplasmic evaginations that appeared to press into the target cell body ( [c, f, and g]). Thus, brain-infiltrating CD8 T cells increased TK cascade phosphorylation induced by TCR signaling (i.e., Lck and ZAP-70), indicating that the CD8 T cells were activated through interaction with antigenic peptides on MHC-I expressed on astrocytes (). Polarization of phosphorylated TKs at contacts between CD8 T cells and target astrocytes strongly implied the visualization of immunological synapse formation in progress.
To determine whether mature immunological synapses containing c-SMAC and p-SMAC develop at the CD8 T cell–astrocyte junctions previously shown to contain polarized tyrosine kinases, we studied the distribution of LFA-1 and TCR on T cells participating potentially in the formation of immunological synapses; we used TK as a marker of virally infected cells. The analysis of LFA-1 and TCR expression in T cells not in contact with infected cells showed a homogeneous, nonpolarized distribution (). The distribution of LFA-1 and TCR at the T cell membrane closely apposed to infected target astrocytes clearly indicated the formation of p-SMAC (LFA-1 rich, TCR poor) and c-SMAC (LFA-1 poor, TCR rich) in vivo ( and ). This indicates that T cell–astrocyte junctions mature to form proper immunological synapses with SMAC formation, concomitantly with the influx of CD8 T cells into the brain, and preceding the clearance of virally infected astrocytes.
The optical images obtained with the confocal microscope were further analyzed using custom-made three-dimensional reconstruction software. Images were α blended to perform a three-dimensional reconstruction from the two-dimensional serial images. This model was rotated in three dimensions, and reconstructions were used to examine in detail the microanatomy of mature brain immunological synapses at the interface plane (–
).
At the interface plane, the p-SMAC is characterized by an LFA-1–immunoreactive ring (–
) surrounding a central area of increased TCR immunoreactivity (–
) and low or absent LFA-1 (, , and ). Note that in most cases LFA-1 immunoreactivity increases within the p-SMAC (), whereas in others there is a drop in c-SMAC LFA-1 (); both distributions provide a p-SMAC–rich/c-SMAC–poor LFA-1 pattern. Finally, in those T cells forming mature immunological synapses, we observed that T cells' nuclei displayed a polarized notch open toward the immunological synapse, indicating that the whole structure of T cells polarized toward the immunological synapse ().
In this study, we have shown that CD8 T cells form typical immunological synapses in vivo displaying p- and c-SMAC at the interface with virally infected astrocytes, preceding and during the clearance of infected cells from the brain. We do so in a model of immune-mediated clearance of infected astrocytes, and demonstrate that both CD8 T cells and CD4 T cells are necessary parts of the effector arm of the immune response, possibly as CD4 Th1 cells. However, detailed morphological analysis demonstrated that CD4 T cells remain circumscribed to the perivascular compartment, whereas CD8 T cells enter the brain parenchyma and form close anatomical contacts with infected cells. A likely explanation is that CD4 T cells may aid in the entry of CD8 T cells into the brain; this is similar to the strategies adopted by both CD4 and CD8 T cells during the clearance of MHV from the brain ().
Although both cell types peak at an early time point in the lymph nodes, it is likely that differential chemokine and adhesion molecule expression by CD4 T cells determines their delayed migration to the CNS and their selective perivascular accumulation when compared with CD8 T cells (–). The delayed peak of intracranial CD4 T cells may serve to facilitate the entry of macrophages into the CNS ().
Initial TCR activation leads to the specific stimulations of the TK signaling pathway; e.g., phosphorylated-Lck and phosphorylated-ZAP-70. These become polarized to areas of close membrane apposition with target cells where mature immunological synapses will form later. Membrane junctions between CD8 T cells with target astrocytes display complex three-dimensional morphology; at times, T cells' membranes completely surrounds individual processes of target-infected cells, or even insinuate their cell body directly into the target astrocyte's soma.
Our experiments demonstrate that CD8 T cells establish SMAC at immunological synapses with infected brain astrocytes that express MHC-I. The data shown herein demonstrate that the mature immunological synapses between T cells–APCs in vivo precede and mediate the clearing of virally infected astrocytes. In addition, we determined that nuclei of T cells formed an open arch toward the immunological synapse, a finding compatible with the known polarization of the microtubule organization center and Golgi apparatus of T cells toward the immunological synapse ().
The previous use of homogeneous populations of cloned T cells, APCs, and time-lapse confocal microscopy has allowed the detailed characterization of the kinetics of T cell–APC interaction and immunological synapse assembly and disassembly in culture (, –). Previous studies of T cell–APC interactions during in vivo immune responses could not demonstrate the formation of p- and c-SMAC, as the result of limited resolution of microscopical techniques used (, ). Multiphoton laser scanning microscopy and confocal microscopy and other advanced imaging methodologies have made major contributions to our dynamic understanding of T cell–APC interactions in lymph nodes and other tissues in vivo (). Although Kawakami et al. () used an ex vivo model to study the influx into the spinal cord of antimyelin CD4 T cells and McGavern et al. () studied entry of CD8 T cells into the meninges of lymphocytic choriomeningitis virus (LCMV)–infected animals, the anatomical resolution of these studies precluded the morphological identification of SMAC formation in vivo (, ) (for a discussion on the status of visualization of immunological synapses in vivo, see reference , Section 7, page 407).
That SMACs indeed form at immunological synapses during in vivo immune responses within the brain parenchyma during clearing of viral infections has hereby been demonstrated. Even if potential contacts between T cells and local APCs were described, the techniques used were unable to demonstrate the existence of SMAC formation during natural immune responses in the brain in vivo (), thus not resolving the existence of SMAC formation as part of in vivo immunological synapse formation.
We believe that the following factors aided in our capacity to detect SMAC formation in vivo in the context of a model antiviral immune response. First, we used an established model in which the kinetics of T cell influx contact with potential targets and elimination of transduced cells were all characterized in much detail. This allowed us to look for formation of immunological synapses at the time of peak T cell entry, but preceding the loss of transduced cells. Second, we used a replication-defective virus expressing a marker gene. This allowed us to identify potential target cells by their expression of a marker gene, without viral replication compromising the survival of infected cells. Third, we optimized the perfusion of animals and our immunocytochemical protocol, in such a way to achieve best preservation of cellular structures through careful perfusion of experimental animals, and full and homogenous antibody penetration throughout the 50-μm-thick vibratome section, through careful improvements to the immunocytochemical protocols used. This allowed us to unravel the three-dimensional structure of interactions between T cells and the complex morphology of target brain astrocytes, and uncover sites of membrane apposition that would not have been found in thinner sections, or in the absence of complete antibody penetration. Fourth, we studied simultaneously the distribution of the essential markers that characterize SMAC formation in immunological synapses, and markers of the target cells, in combination within single sections using four-color immunostaining. Finally, the custom-made three-dimensional reconstruction software for confocal images allowed us to rotate cells in close anatomical appositions and observe the distribution of the several markers at different optical planes in a way that provides a complete picture of the distribution of immunological synaptic proteins in three full dimensions. This resulted in the typical “bull's eye” images of the mature immunological synaptic interfaces illustrated in .
Although the functional consequences of molecular segregation in the immunological synapse remain under investigation (, ), recent work has demonstrated that immunological synapses serve to channel cytokines and effector molecules toward target cells (, ). That mature immunological synapses were found in the brain preceding the clearance of viral infected cells, suggests that the presence of SMAC-containing immunological synapses in a physiological context in vivo may be necessary for clearance of virally infected cells to occur. However, the ultimate proof of this hypothesis will have to wait the development of compounds that selectively inhibit immunological synapse formation.
The demonstration of mature immunological synapses in vivo during antiviral immune responses will allow further experimental exploration of immunological synaptic function during normal and pathological immune responses in vivo. Furthermore, this work should help contribute to settling the controversy over the existence and functional significance of mature immunological synapses in vivo during antiviral immune responses. In summary, we propose that mature, SMAC-containing immunological synapses are the anatomical substrate that mediates the complex sequence of activation and effector function of T cells in vivo, as CD8 T cells clear virally infected cells from the CNS.
Adult male Sprague-Dawley rats (250 g body weight) (Charles River) were used according to Cedars- Sinai Medical Center's Institutional Animal Care and Use Committee–approved protocols. Adenoviruses used in this study were first-generation E1/E3-deleted recombinant adenovirus vectors based on adenovirus type 5. The construction of RAdTK (expressing herpes simplex virus type I thymidine kinase, HSV1-TK) and RAdHPRT (expressing hypoxanthine-guanine phosphoribosyl-transferase), all contain the hCMV promoter and have all been described in detail elsewhere (). Animals were injected unilaterally in the left striatum with 10 infectious units (i.u.) of RAdTK in a volume of 1 μl and immunized 30 d later with 5 × 10 infectious units of RAdHPRT injected subcutaneously. Animals were killed for analysis at different time points after immunization as described in more detail in the following sections.
75 animals were injected with 10 i.u. of RAdTK into the striatum at day 0. 1 mo later, rats were anaesthetized briefly and immunized subcutaneously in the back with 100 μl of either sterile saline ( = 15) or 5 × 10 i.u. of RAdHPRT ( = 60). 3 d after systemic immunization against adenovirus, one group of animals ( = 15) was injected weekly with 0.5 mg of OX8 monoclonal antibody i.p. to deplete CD8 T cells. Another group ( = 15) was injected i.p. with 1 mg of OX34 monoclonal antibody every 2 wk to deplete CD4 T cells, and two groups of animals were injected with mouse monoclonal irrelevant isotype antibodies as appropriate controls. The number of CD4 and CD8 T cells were quantified in draining cervical lymph nodes (Fig. S1). Results of the depletion studies are shown in . 7, 14, and 30 d after the immunization, five animals from each experimental group were killed via anesthetic overdose, transcardially perfused with 200–500 ml of oxygenated Tyrode solution, and spleen and cervical lymph nodes were removed for flow cytometry. Immediately afterward, animals were perfused/fixed with 4% paraformaldehyde to fix the brain. Brains were postfixed in 4% paraformaldehyde for up to 48 h, after which they were washed in phosphate buffer, cut into smaller tissue pieces, and sectioned as described in the following paragraph. This procedure provides excellent quality preservation of brain tissue for further analysis. For the CD8 depletion studies, postfixed brains were sectioned on a vibratome (Leica Instruments) at 50-μm section thickness. In the CD4 depletion studies, brains were cryoprotected in 20% sucrose and 16-μm sections were cut on the cryostat (Leica Instruments).
50-μm coronal brain sections were cut serially through the striatum on a Leica vibratome, and immunofluorescence or DAB detection was performed as described previously (), using the following primary antibodies recognizing: CD8 (1:500, mouse, Serotec), CD4 (1:100, mouse, Serotec), TK (1:10,000, rabbit, custom made), NeuN (1:1,000, mouse, Chemicon), GFAP (1:500, guinea pig, Advanced Immunochemical), phosphorylated Lck (1:50, rabbit, Cell Signaling), LFA-1 (1:500, mouse, IgG2a, BD Biosciences), TCR (1:100, mouse, IgG, BD Biosciences), phosphorylated ZAP-70 (1:100, rabbit, Cell Signaling), and MHC-I (1:1,000, mouse, Serotec). Sections were examined using a Leica DMIRE2 confocal microscope (Leica Microsystems). Three-dimensional reconstructions to allow rotation of the images were rendered with α-blending software (custom made by K. Wawrowsky). Striatal sections from 25 animals were screened and analyzed in their whole extent searching for T cells interactions; a total of at least 60 immunological synapses (a likely underestimation of the total number of synapses present in the brain) at various stages of development were recorded and analyzed in detail.
To determine the approximate number of activated T cells over the total number of T cells, the expression of phosphorylated ZAP-70, a marker of T cell activation after TCR engagement, and thus, a close surrogate of potential immunological synaptic engagement, was quantified. This analysis indicated that 75.4% of all CD8 T cells within the injected striatum express phosphorylated ZAP-70. This provides a quantitative estimate of the potential frequency of immunological synapses being established by T cells.
Three-dimensional reconstructions where generated with a custom-made software. Images were α-blended to perform a three-dimensional reconstruction from the two-dimensional serial images. The images were rotated in three-dimensional perspective to get the optical plane of the interface. The criteria of choosing the interface are illustrated in detail in . Artistic renderings (made by C. Barcia) were produced based on the three- dimensional analysis for and .
Adenovirus neutralizing antibody titers were measured in serum samples as described by us (, ) (Fig. S1).
12 rats were injected bilaterally into the striatum with 10 infectious units of RAdTK. 1 mo later, six animals were control immunized with saline, and six animals were immunized intradermally with RAdHPRT. Rats were killed 60 d after the immunization by neck dislocation and decapitation, after an anesthetic overdose. Brains were removed, cut into five blocks (left and right forebrain, left and right midbrain, and cerebellum), snap frozen in liquid nitrogen, and stored at −80°C. Frozen blocks of the left or right forebrain (containing the striatum) were ground to a fine powder under liquid nitrogen. DNA was extracted from 50 mg of ground tissue using spin columns for DNA purification (QIAGEN). 250 ng of DNA, or a quantity of cDNA corresponding to 100 ng of total RNA was amplified by PCR using TKf 5′-CGAGCCGATGAC TTACTGGC-3′ and TKr 5′-CCCCGGCGGATATCTCAC-3′ primers and the FAM-labeled TKp 5′-TACACCCAACACCGCCTCGACC-3′ TaqMan probe (PerkinElmer). RNA control was performed using two primers and a VIC-labeled 18S ribosomal probe (PerkinElmer). The TK copy number standard curve was obtained using dilutions of the previously described EpTK plasmid. The final concentration was 0.2 μM for each primer, probe 0.2 μM, dNTP 0.2 mM (except for dUTP 0.4mM), MgCl 5 mM, UNG 0.5U in PCR buffer in the presence of 1.25 U of AmpliTaq Gold polymerase (PCR core reagents, PerkinElmer). DNA determinations were made in triplicate. Amplifications used a ABI Prism 7700 sequence detection system (PerkinElmer) with an initial step at 50°C for 2 min, 95°C for 10 min, followed by 40 cycles of the following: denaturation at 95°C for 15 s and annealing/extension at 60°C for 1 min. Raw data were analyzed using the SDS v1.6.3. software (PerkinElmer).
Viability data were expressed as mean ± SEM and evaluated by two- or one-way analysis of variance (followed by Dunnet or Tukey multiple comparisons tests). Differences were considered significant if P < 0.05. When significance testing using analysis of variance was not applicable, the Kruskal-Wallis nonparametric test was used instead.
Fig. S1 illustrates neutralizing antiadenoviral antibody titers in immunized and nonimmunized rats, and levels of CD4 and CD8 T cells in cervical lymph nodes of naive and immunized rats. Fig. S2 illustrates the effects of specific depleting antibodies on levels of CD8 and CD4 T cells in the spleen. The supplemental Materials and methods section includes detailed descriptions of immunocytochemistry and its stereological quantification. Online supplemental material is available at . |
Model systems are vital for the study of disease and the development of new therapeutic approaches. The dog was not immediately recognized as a model for hereditary diseases and few genetic-oriented investigations were carried out until the mid-1990s. Since that time, the canine genetics research community has made significant strides, producing dense linkage and radiation hybrid maps, oligo-based microarrays, SNP arrays, and, most importantly, the sequence of the canine genome at 7.6X coverage (Breen et al. ; Clark et al. ; Guyon et al. ; Linblad-Toh ; Linblad-Toh et al. ).
The dog offers many of the same advantages of other small animal models. For example, homogeneous populations exist in each of the hundreds of pure breeds, and pedigrees can be easily established in rapid fashion. One advantage that other model systems do not have is that for the dog, pet populations can often be utilized, thereby often eliminating the need for establishment of colonies. Dogs possess other characteristics that are not found in traditional rodent models in that they receive exceptional medical care, have comparable organ sizes (to humans), and generally cohabitate with their human owners, minimizing different environmental effects (Ostrander et al. ). The last issue is of particular interest. Specifically, because the dog does live with us, it is exposed to the same environment. Of course, the dog may react differently to such influences/stress than does the human, but living in the same environment is more advantageous when compared to environments in which classic laboratory research animals are maintained because those environments are far different from the ones inhabited by humans. Thus, when modeling the causes and pathogenesis of human hereditary diseases, any environment-gene interactions are likely better studied in an animal that lives in the same environment.
Perhaps the most exciting feature of the canine model is that 220 naturally occurring disease phenotypes are potential models for various human hereditary diseases (Online Mendelian Inheritance in Animals ; ).
While the mouse is indisputably a fundamental resource for the study of human hereditary diseases, the canine model offers the opportunity to gain knowledge in areas for which the mouse is deficient. For example, genetically altered mice are not available for every disease of interest and alternative models must be identified in these instances. There are more than 450 canine hereditary diseases that provide naturally occurring models in which to study diseases (Ostrander and Giniger ). Among these are diseases transmitted in X-linked, autosomal recessive, and autosomal dominant fashions. Also, there are diseases for which canine models were identified before the development of murine models and were used to investigate pathogenesis and treatment regimens. An example of this is hereditary nephropathy (Kashtan ).
Research pertaining to spontaneous phenotypes of the dog has revealed genes and pathways novel to diseases. A prime example of this is the work on narcolepsy. In the 1990s, identification of the genetic basis of narcolepsy in the dog led investigators to a pathway not previously known to be involved in the disorder (Lin et al. ). Another example is the recent identification of a novel gene that causes retinal degeneration in the dog. It was subsequently determined that an identical mutation in the homologous gene was also responsible for a similar phenotype in a human patient (Zangerl et al. ). While these are hallmark examples of the utility of the canine system, its full potential has yet to be realized because, to date, only a fraction of all hereditary traits have been characterized at the molecular level ().
Phenotypes resulting from induced models of disease are not always clinically equivalent to those observed in humans. For example, a knockout mouse with muscular dystrophy shows muscular weakness but not the continued wasting that is found in human patients (Tanabe et al. ). Such incongruities may diminish the usefulness of the model, specifically with regard to testing of possible treatments. In the last decade spontaneous canine models have been instrumental in the development of molecular therapies for human disease; e.g., data obtained from a canine model of hemophilia B led to clinical trials in humans (High ). Presented here is a review of the aforementioned diseases for which examination of canine models has revealed previously unknown genetic bases and/or facilitated development of novel treatment options.
Hereditary nephropathy (HN) is a broad term for certain fatal inherited diseases that result in renal failure. Alport syndrome (AS) is a form of HN in humans caused by defects in the glomerular basement membrane (GBM) (Kashtan ; Tryggvason and Martin ). The only treatments currently available for AS are dialysis and renal transplant. Mutations in the type IV collagen genes cause AS, which is primarily inherited in X-linked (XLAS) and autosomal recessive (ARAS) fashions. There is also a rare autosomal dominant (ADAS) form (Hudson et al. ). XLAS results from mutations in and accounts for 85% of cases, while ARAS and ADAS are caused by mutations in or (Jais et al. ; Knebelmann et al. ; Lemmink et al. ; Martin et al. ; Mochizuki et al. ). Mutations in any of these genes alter the triple helix formed by the COL4A3, COL4A4, and COL4A5 proteins that are necessary for proper GBM formation in the kidney.
Murine models for AS did not exist until 1996 when two transgenic models for ARAS were developed and characterized (Cosgrove et al. ; Miner and Sanes ). A murine model for XLAS was not described until 2004, despite it being the most common genetic form of AS (Rheault et al. ). To date, a murine model for ADAS has not been developed. Naturally occurring HN has been identified in several canine families. The progression of the disease is very similar to AS in humans with the exception of auditory and ophthalmologic abnormalities, which have not been described in the dog. X-linked HN (XLHN) in the dog was identified first in the Samoyed and later in a mixed-breed family (Fig. ) (Jansen et al. ; Lees et al. ). A single-base substitution in exon 35 in the Samoyed and a 10-bp deletion in exon 9 in the mixed-breed dog result in premature stop codons and truncated COL4A5 proteins (Cox et al. ; Zheng et al. ). The English cocker spaniel presents with a renal disease (termed ARHN) similar to ARAS that is caused by a nonsense mutation in exon 3 of (Davidson et al. ). The bull terrier is affected by ADHN but the mutation has not been characterized (Hood et al. ).
Because dialysis and renal transplants are the only treatments available for AS, correction of the defective GBM via gene therapy as a possible remedy is being explored. In an initial experiment, an adenoviral vector containing a human cDNA construct of was successfully expressed in the kidney of normal pigs (Heikkilä et al. ). This study proved that the construct could produce a functional protein capable of trimerizing with COL4A3 and COL4A4 (Heikkilä et al. ). To assess the functionality of this in a diseased system, a canine cDNA construct was designed for use in a canine model of the disease (Harvey et al. ). Because vector delivery methods are complicated in the kidney, studies were carried out in the bladder. In the smooth muscle of the bladder, COL4A5 and COL4A6 form a trimer composed of two transcripts of COL4A5 and one transcript of COL4A6. In XLHN-affected animals, COL4A6 is not deposited into the basement membrane because COL4A5 is not available to form the trimer. Therefore, for gene therapy to be effective, not only must the construct be produced, but the other type IV collagens must also be able to properly trimerize with it. An adenoviral vector containing the canine cDNA of was injected into smooth muscle of the bladder in XLHN-affected Samoyed dogs. Five weeks after injection, expression of both COL4A5 and COL4A6 was found in the basement membranes surrounding the injection site, indicating that COL4A5 was expressed and functional and that COL4A6 was made and deposited (Harvey et al. 2006). These findings are promising for gene therapy of HN and AS.
Narcolepsy is a neurologic condition characterized by excessive daytime sleepiness and cataplexy, the sudden loss of muscle tone (Mignot ). This rare sleep disorder, which affects less than 0.1% of humans, is debilitating and difficult to diagnose (Dauvilliers et al. ). Most cases are sporadic, with familial causes representing fewer than 10% of cases, and twin studies indicate a strong influence of nongenetic factors (Mignot ).
In the 1980s, an association with narcolepsy and specific HLA genes and alleles was identified in several ethnic populations (Juji et al. ; Seignalet and Billiard ). Subsequent studies of these genes have shown that they do not harbor deleterious mutations but rather confer susceptibility to the disease (Maret and Tafti ). Based on this significant association with the major histocompatibility (MHC) system, it has been suggested that narcolepsy may result from an autoimmune reaction to environmental agents (Lin et al. ).
Naturally occurring narcolepsy was first described in the dog in the 1970s (Knecht et al. ). Narcoleptic dogs have clinical signs that parallel those observed in humans and were used as a model to elucidate the genetics underlying the disorder (Hungs et al. ). In the dog, narcolepsy is inherited in an autosomal recessive fashion and, unlike the human, is not associated with the dog leukocyte antigen (DLA) system (Foutz et al. ; Wagner ). In 1999, colonies of Doberman pinchers and Labrador retrievers with narcolepsy were used in positional cloning efforts that identified linkage with the () gene (Lin et al. ). Hypocretin proteins (orexins), discovered in 1998, are neurotransmitters processed from a common precursor, preprohypocretin (Hcrt) (de Lecea et al. ; Sakurai et al. ). Independent mutations in these canine families cause exon-skipping and result in altered proteins. In addition, Hungs et al. () identified a single-base change resulting in an amino acid substitution in Hcrtr-2 that causes narcolepsy in a family of Dachshunds. does not cause narcolepsy in a family of poodles or in 11 individual cases with no family history (Hungs et al. ).
The role of hypocretins in canine narcolepsy, and the simultaneous finding that knockout mice have narcoleptic symptoms (Chemelli et al. ), led to the investigation of hypocretins in humans. and its receptors encoded by and were sequenced in human patients with narcolepsy and a substitution mutation in of a single patient was identified (Peyron et al. ). Although mutations in these genes may be a rare cause of human narcolepsy, expression studies show that hypocretins are important in the etiology of the disorder. Hypocretin concentrations are below average or undetectable in the cerebrospinal fluid of most narcoleptic patients, indicating deficient neurotransmission (Nishino et al. ; Peyron et al. ).
Current treatments address only the symptoms of narcolepsy and have significant side effects (Nishino et al. ). The availability of a naturally occurring canine model has allowed pharmacologic studies to improve treatments for narcoleptic patients (Nishino et al. ). Animal models are currently being used to study the hypocretin system as a target for new therapeutic approaches (Dauvilliers and Tafti ). In addition, the canine models in which the causative factors remain unknown may still be useful for unmasking the genetic and environmental factors that are associated with narcolepsy.
Retinitis pigmentosa (RP) is a group of hereditary disorders characterized by progressive retinal degeneration, eventual night blindness, loss of peripheral vision, and often complete blindness. RP affects 1 in 4000 people and may be inherited as an autosomal dominant, autosomal recessive, or X-linked trait (Hartong et al. ). More than 45 genes, accounting for 60% of all cases, have been implicated in RP (Hartong et al. ).
Progressive retinal atrophy (PRA) is a group of hereditary diseases of the dog that are phenotypically and molecularly similar to RP. Progressive rod-cone degeneration (prcd) is an autosomal recessive, late-onset form of PRA. In 1998, the locus was mapped to the centromeric end of CFA9 using classical linkage analysis (Acland et al. ). Fortuitously, prcd in multiple breeds results from allelic or identical mutations in the gene. This allowed for multiple breeds to be used in a large-scale linkage disequilibrium (LD) approach, which further defined the interval that harbors the locus (Goldstein et al. ). A retinal EST library was instrumental in the identification of novel candidate genes in this region (Zangerl et al. , ). A single G-to-A transition in a gene of unknown function, termed , was found to cause prcd in at least 18 breeds (Zangerl et al. ).
Upon discovery of the canine mutation, human patients with inherited retinal disorders for which known causative mutations had been excluded were screened for mutations in the gene. A woman from Bangladesh having an autosomal recessive form of RP was found to harbor a homozygous mutation in the gene (Zangerl et al. ). Interestingly, the mutation is identical at the genetic and protein level to the mutation causing prcd in the dog (Zangerl et al. ).
The similarities between humans and dogs can be exploited to develop treatments for retinal degenerative diseases, with significant strides in this field having already been made. In 2001, gene therapy restored vision to dogs with severe retinal degeneration caused by a homozygous, 4-bp deletion in (Acland et al. ). The dose efficacy and safety data obtained from the successful use of the canine model have led to gene therapy studies in nonhuman primates and the consideration of human trials (Jacobson et al. ).
Hemophilia B is a recessive bleeding disorder that results from mutations in the () gene on the X chromosome. FIX, synthesized by hepatocytes, is an essential part of the blood coagulation cascade. Clotting factor deficiencies result in bleeding into joints, soft tissue, and muscles. Such bleeding may occur spontaneously or be triggered by a minor injury. Hemophilia B is estimated to occur in 1 in 30,000 males and is both clinically and molecularly heterogeneous. Approximately 1000 unique mutations causing hemophilia B have been reported in humans (Green et al. ).
Canine hemophilia B is highly similar to the human disease and is well studied. In 1989, Evans et al. () published the coding sequence of canine . The same group also identified the first known mutation to cause hemophilia B in the dog, a missense mutation resulting in the complete absence of detectable protein (Evans et al. ). Since then, multiple cases have been described in different breeds and distinct mutations for five of these have been reported (Table ) (Brooks et al. , ; Gu et al. ; Mauser et al. ).
The standard treatment for hemophilia B is intravenous infusion of FIX concentrates to prevent or treat bleeding episodes. While treatment is effective and generally safe, it is expensive and inconvenient (Lillicrap et al. ). Molecular therapies for hemophilia B are being investigated not only to provide patients with treatment options but also to evaluate the overall efficacy of such approaches. The factors that make hemophilia a superior model for assessment of genetic intervention strategies are tissue-specific gene expression is not necessary, nominal increases in clotting factor levels will result in significant phenotypic improvements, measurement of clotting factor levels can be achieved through simple blood tests, and naturally occurring canine models are available (Lillicrap et al. ).
A colony of dogs with hemophilia B due to a missense mutation first characterized by Evans et al. (), has been used in numerous pioneering gene therapy studies (Kay et al. , ; Snyder et al. ; Wang et al. ). The first of these utilized a retroviral vector containing cDNA and resulted in long-term expression of low levels of FIX (Kay et al. ). A subsequent study used a recombinant adenoviral vector and achieved short-term expression of therapeutic levels of FIX (Kay et al. ). To attain both long-term and therapeutic levels of FIX expression in these dogs, researchers found success by using adeno-associated viral (AAV) vectors (Herzog et al. ; Snyder et al. ; Wang et al. ). An AAV vector was also later used to correct a severe hemophilia B phenotype in dogs with a null mutation (Mount et al. ).
The data obtained using the AAV vector in these canine studies provided the proof of principle necessary to move forward with hemophilia B gene therapy trials in humans (High ). The dog has also been instrumental in advancing molecular therapy approaches for hemophilia A, which results from factor VIII deficiency (Lillicrap et al. ). Hemophilia A is significantly more common than hemophilia B, but gene therapy studies have been impeded by large size of the gene (Kay and High ).
The most common and severe muscular disorder in humans is Duchenne muscular dystrophy (DMD). DMD is an X-linked disorder that results in muscle degeneration and death around the age of 20. It affects approximately 1 in 3500 males and there are currently no effective treatments available. A naturally occurring form of DMD has been described in the golden retriever (Kornegay et al. ). Golden retriever muscular dystrophy (GRMD) is characterized by elevated serum creatinine kinase activity, progressive muscle atrophy and necrosis, and regeneration by fibrotic and adipose tissues. Affected dogs develop clinical signs in 8-10 weeks (Kornegay et al. ).
DMD is caused by a defective dystrophin gene, which codes for a cytoskeletal protein responsible for stabilizing the sarcolemma (Hoffman et al. ). Northern and Western blots using human probes failed to detect dystrophin transcripts or proteins in muscle tissue from GRMD-affected dogs (Cooper et al. ). Sequence analysis of the dystrophin gene revealed an A-to-G transition in the exon 7 splice acceptor of affected dogs (Sharp et al. ). This mutation causes either the deletion of exon 7 or the use of an alternative splice site 5 bp downstream; both result in a reading frame shift and a truncated transcript (Dell’Angola et al. ; Sharp et al. ).
The frequency and severity of DMD has fueled interest in the development of gene therapies for patients. Significant advancements have come from a murine model, but greater similarities in disease progression make the dog a more attractive model (Foster et al. ; Tanabe et al. ). Initial gene therapy studies in the dog focused on the gene and were promising (Howell et al. ). An alternative approach involved upregulation of , a gene functionally and structurally similar to , but not foreign to DMD patients (Cerletti et al. ). Delivery of mini-utrophin transcripts via an adenoviral vector mitigated the dystrophic phenotype in the muscles of GRMD dogs; however, slight immunologic reactions to the vector and transgene occurred (Cerletti et al. ). An additional method involved modified antisense oligonucleotides (AOs) that cause exon-skipping. By changing the splicing pattern, AOs can cause a mutated exon to be removed from the pre-mRNA, leading to a functional protein. McClorey et al. () successfully used AOs to restore dystrophin expression in dogs.
Human gene therapy studies have been successful as well, with several types of vectors used to deliver dystrophin to dystrophic muscles. AAV vectors have been problematic because of their limited carrying capacity and the large size of the dystrophin gene. The use of microdystrophin (a truncated but functional version of the dystrophin gene) has shown promise and initial trials are underway (Foster et al. ). In addition, with the success of AO treatment in the dog, human phase 1 clinical trials have been initiated (Foster et al. ; , ).
In an effort to eliminate immune reactions against vectors and/or the gene itself, researchers investigated the use of stem cells to treat GRMD. Hematopoetic stem cells have proven to be effective in muscle regeneration in the murine model (Gussoni et al. ). Unfortunately, hematopoetic stem cells from normal littermates did not cause muscle regeneration in affected dogs (Dell’Angola et al. ). Vessel-associated stem cells, called mesoangioblasts, were also studied. Mesoangioblasts were successfully transplanted and expressed dystrophin, allowing recovery of muscle use (Sampaolesi et al. ). Donor wild-type mesoangioblasts were found to be more effective than genetically corrected autologous mesoangioblast cells (Sampaolesi et al. ). All dogs treated with wild-type cells displayed initial mobility improvements and one dog was still walking five months after cessation of treatments (Sampaolesi et al. ).
Presented here are examples of studies that were critical to identification and treatment of genetically simple diseases that affect the human and dog. However, the genetics of complex diseases are more difficult to assess—in the dog and human. Nevertheless, with new genomic tools/resources now available for study of the dog, workers have now begun analyses of complex diseases such as cancers, cardiovascular diseases (e.g., cardiomyopathy), and neurologic diseases (e.g., epilepsy). In addition, the dog is being used to assess the genetics of morphologic development and behavior due to the unique physical and behavioral traits that characterize individual breeds. The hypothesis for such lines of investigation is simple: while there likely are additional factors that influence behavior, morphology, and progression of diseases in the dog and human, the major genes influencing these may very well be the same.
One complex disease for which data are available from the dog is canine hip dysplasia (CHD) or degenerative joint disease, the major orthopedic disease of the dog. This is a painful and crippling disease that has a counterpart in the human termed developmental dislocation of the hip. Researchers are exploiting the natural occurrence of CHD in both pet populations and designed outcrossed pedigrees in an attempt to delimit contributory genetic components to this disease (Chase et al. ; Todhunter et al. ; Tsai and Murphy ). To date, two QTLs have been identified in the Portuguese water dog (PWD) on CFA01 that are associated with joint laxity as measured by the Norberg angle. Interestingly, one of the QTLs is associated with joint laxity in the right hip while the other is associated with the left hip (Chase et al. ). The same PWD population was used recently to identify the insulin-like growth factor 1 gene () as a determinant for skeletal size in dogs (Sutter et al. ). This is an important finding because understanding the genetics of growth and regulation may provide insight into complex diseases such as cancer and hip dysplasia.
The diseases discussed in this article highlight the importance of the dog to biomedical research, particularly the study of hereditary diseases. Perhaps most unique about the use of the dog as a model is this: study of those hereditary diseases common to the dog and human allows both to benefit as opposed to one serving merely as a model for the other. A paucity of genetic tools with which to study the canine genome previously prohibited researchers from exploiting the canine model. Thus, the majority of research in the dog was guided by our knowledge of the disease in the human. With the necessary resources now available, discoveries in the dog are being used to define causative genes and pathways and, importantly, develop new treatment regimens for the human. |
Genetics is at the dawn of a new era with maturing technologies that enable low-cost, high-throughput genotyping of hundreds of thousands of DNA markers that in turn can be tested for association to complex traits of interest like disease and drug response. A number of studies have already leveraged the availability of such technologies to identify polymorphisms in genes that associate with diseases like age-related macular degeneration (Edwards et al. ; Haines ; Klein ), diabetes (Grant ; Sladek ), and obesity (Herbert ), to name just a few. In addition, there are scores of similar genome-wide association studies that are ongoing and that promise to deliver scores of genes that harbor variations that associate with diseases like obesity and diabetes. While these types of genetic discoveries provide a peek into pathways that underlie disease, they are usually devoid of context, so that elucidating the functional role such genes play in disease can take years, or even decades, as has been the case for ApoE, an Alzheimer’s susceptibility gene identified nearly 15 years ago (Peacock et al. ).
Information that defines how variations in DNA that associate with disease actually impact the complex physiologic processes underlying disease flows through transcriptional and other molecular, cellular, tissue, and organism networks (Fig. ). In the past the ability to comprehensively assess intermediate phenotypes that comprise the hierarchy of networks that drive disease was not possible. However, today DNA microarrays have radically changed the way we study genes, enabling a more comprehensive look at the role they play in everything from the regulation of normal cellular processes to complex diseases like obesity and diabetes. In their typical use, microarrays allow researchers to screen thousands of genes for differences in expression or differences in how genes are connected in molecular networks (Schadt and Lum ) between experimental conditions of interest. These data are often used to discover genes that differ between normal and disease-associated tissue, to model and predict continuous or binary measures, to predict patient survival, and to classify disease or tumor subtypes. Because gene expression levels in a given sample are measured simultaneously, researchers are able to identify genes whose expression levels are correlated, implying an association under specific conditions or more generally.
Integrating genetic and functional genomic data can provide a path to inferring causal associations between genes and disease. In the past, causal associations between genes and traits have been investigated using time series experiments, gene knockouts or transgenics that overexpress a gene of interest, RNAi-based knockdown or viral-mediated overexpression of genes of interest, and chemical activation or inhibition of genes of interest. A more systematic and arguably relevant source of perturbation to make such inferences regarding genes and disease are DNA polymorphisms, where gene expression and other molecular phenotypes in a number of species have been shown to be significantly heritable and at least partially under the control of specific genetic loci (Brem et al. ; DeCook et al. ; Hubner et al. ; Jin et al. ; Klose et al. ; Monks et al. ; Morley et al. ; Oleksiak et al. ; Schadt et al. ; Stranger et al. ). By examining the effects that naturally occurring variations in DNA have on variations in gene expression traits in human or experimental populations, other phenotypes (including disease) can be examined with respect to these same DNA variations and ultimately ordered with respect to genes to infer causal control (Fig. ) (Kulp and Jagalur ; Lum et al. ; Mehrabian et al. ; Schadt et al. ). The power of this integrative genomics strategy rests in the molecular processes that transcribe DNA into RNA and then RNA into protein, so that information on how variations in DNA impact complex physiologic processes often flows directly through transcriptional networks. As a result, integrating DNA variation, transcription, and phenotypic data has the potential to enhance identification of the associations between DNA variation and disease and characterize those parts of the molecular networks that drive disease.
Here we review different approaches for integrating expression quantitative trait loci (eQTLs), expression, and clinical data to infer causal relationships among gene expression traits and between expression and disease traits. We further review methods to integrate these data in a more comprehensive manner by constructing coexpression gene networks that leverage pairwise gene interaction data to represent more general relationships. This type of network provides a useful construct for characterizing the topologic properties of biological networks and for partitioning such networks into functional units (modules) that underlie complex phenotypes like disease. However, these networks are, by design, undirected and so do not explicitly capture causal relationships among genes. To infer gene networks that capture causal information, we review Bayesian network reconstruction algorithms that, like the methods operating on only two or three expression traits and/or clinical traits mentioned above, integrate eQTLs, expression, and clinical phenotype data to reconstruct whole-gene networks capable of representing direction along the edges of the network. Here, directionality among the edges corresponds to causal relationships among genes and between genes and clinical phenotypes related to disease. These emerging high-dimensional data analysis approaches that integrate large-scale data from multiple sources represent the first steps in statistical genetics, moving away from considering one trait at a time and toward operating in a network context. Evolving statistical procedures that operate on networks will be critical to extracting information related to complex phenotypes like disease as research goes beyond the single-gene focus. The early successes achieved with some of the methods described herein suggest that these more integrative genomics approaches to dissecting disease traits will significantly enhance the identification of key drivers of disease beyond what could be achieved by genetics alone.
Gene transcripts have been identified that are associated with complex disease phenotypes (Karp et al. ; Schadt et al. ), are alternatively spliced (Johnson et al. ), elucidate novel gene structures (Mural et al. ; Schadt et al. ; Shoemaker et al. ), can serve as biomarkers of disease or drug response (DePrimo et al. ), lead to the identification of disease subtypes (Mootha et al. ; Schadt et al. ; van’t Veer et al. ), and elucidate mechanisms of drug toxicity (Waring et al. ). Changes in gene expression often reflect changes in a gene’s activity and the impact a gene has on different phenotypes. Because gene expression is a quantitative trait, linkage and association methods can be directly applied to such traits to identify genetic loci that control them. In turn, genetic loci that control for expression traits may also associate with higher-order phenotypes affected by expression changes in the gene of interest, providing a path to directly identify genes controlling for phenotypes of interest. Therefore, identifying the heritable traits and the extent of their genetic variability provides insight about the evolutionary forces contributing to the changes in expression that associate with biological processes that underlie diseases like obesity and diabetes, beyond what can be gained by looking at the transcript abundance data alone.
It is now well established that gene expression is a significantly heritable trait (Alberts et al. ; Brem et al. ; Chesler et al. ; Cheung et al. ; Jansen and Nap ; Monks et al. ; Morley et al. ; Petretto et al. , ; Schadt et al. , ). If a gene expression trait is highly correlated with a disease trait of interest, and if the corresponding gene physically resides in a region of the genome that is associated with the disease trait, then knowing that the expression trait is also genetically linked to a region coincident with its physical location provides an objective and direct path to identify candidate causal genes for the disease trait (Alberts et al. ; Brem et al. ; Chesler et al. ; Cheung et al. ; Jansen and Nap ; Monks et al. ; Morley et al. ; Petretto et al. , ; Schadt et al. , ). The genetic information therefore enables the dissection of the covariance structure for two traits of interest into genetic and nongenetic components, and the genetic component can then be leveraged to support whether an expression and disease trait are related in a causal, reactive, or independent manner (with respect to the expression trait). Elucidating causal relationships is possible in this setting given the unambiguous flow of information from changes in DNA to changes in RNA and protein function (Fig. ). That is, given that two traits are linked to the same DNA locus and a few important simplifying assumptions, there are a limited number of ways in which these two traits can be related with respect to a given locus (GuhaThakurta et al. ; Schadt ; Schadt et al. ), whereas in the absence of such genetic information, many indistinguishable relationships would be possible, so that additional data would be required to establish the correct relationships.
Leveraging DNA variation information to reconstruct gene networks supposes that we are able to systematically identify genetic loci that at least partially control transcript abundances for genes of interest. This of course is straightforward given that transcript abundance or gene expression traits are quantitative measures that can be analyzed like any other quantitative trait in a genetics context. However, the difficulty in analysis and interpretation comes with the large number of traits examined. Microarrays are capable of monitoring tens of thousands (or hundreds of thousands) of transcripts simultaneously. Therefore, methods to compute eQTLs must consider computational tractability given the need to run the analyses potentially millions of times. In addition, significance thresholds must take into account multiple testing. Multiple testing issues relate not only to the number of transcripts tested but also to the number of markers or proportion of the genome tested. However, the strong correlation structure that exists among many of the expression traits monitored in a segregating population can be leveraged to enhance the power to detect relationships among genes.
A number of methods have been developed and applied to gene expression traits in segregating populations to identify eQTLs and to establish relationships among genes and between genes and disease traits, where multiple traits at a time can be considered. Typical approaches to the joint analysis of genetic traits involve mapping each gene expression trait individually and inferring the genetic correlation between pairs or sets of expression traits based on pairwise Pearson correlation, eQTL overlaps, and/or tests for pleiotropy. Using a family-based sample, Monks et al. () estimated the genetic correlation between pairs of traits using a bivariate variance-component-based segregation analysis and showed that the genetic correlation was better able to distinguish clusters of genes in pathways than correlations based on the observed expression traits. This type of method can be extended to perform bivariate and multivariate QTL analyses, which can be more highly powered to detect QTLs when traits are correlated. Clusters of correlated gene expression traits can often contain hundreds or thousands of genes, which would be computationally prohibitive in a joint analysis. Kendziorski et al. () approached this problem in a different way by employing a Bayesian mixture model to exploit the increased information from the joint mapping of correlated gene expression traits, which is computationally tractable for large sets of genes. Instead of doing a linkage scan by computing LOD scores at positions along the genome, Kendziorski et al. () computed the posterior probability that a particular gene expression trait maps to marker for each marker, as well as the posterior probability that the trait maps nowhere in the genome. Nonlinkage in this setting is declared for a transcript if the posterior probability of nonlinkage exceeds a threshold that bounds the posterior expected false discovery rate (FDR). One benefit to this approach is that it controls false discovery for the number of expression traits being tested, whereas assessing the appropriate significance cutoffs in single-transcript linkage analysis often requires data permutation analyses. The drawback of this method is that it assumes that linkage occurs at either one or none of the markers tested and it lacks a well-defined method for the case when multiple eQTLs control an expression trait.
In a study of inbred strain crosses, the only valid way of estimating the extent of genetic control of a given trait is to explicitly model each eQTL, including any epistatic interactions if they exist. Brem and Kruglyak () showed that epistatic interactions were prevalent in the gene expression levels in yeast, and similar suggestions have been made in other species as well (Schadt et al. ), but more definitive studies are needed to characterize the extent of epistasis among eQTLs in these other species. In the absence of epistasis, the genetic contribution for each transcript has been estimated by summing contributions for each individual eQTL, assuming that little or no allelic association exists between the eQTLs. In the presence of epistasis, however, this practice cannot yield a valid estimate, and multilocus models are instead required to obtain valid estimates. In addition, multilocus modeling can identify loci contributing to expression traits that would have been missed in single-locus eQTL scans (Brem and Kruglyak ; Storey et al. ).
While understanding the mechanisms of RNA expression is in itself important for understanding biological processes, the ultimate use of this information is identifying the relationship between variation in expression levels and disease phenotypes in an organism of interest. Microarray experiments are commonly used to explore differential expression between disease and normal tissue samples or between samples from different disease subtypes. These studies are designed to detect association between gene expression and disease-associated traits, which in turn can lead to the identification of biomarkers of disease or disease subtypes. However, in the absence of supporting experimental data, these data alone are not able to distinguish genes that drive disease from those that respond. As discussed above, eQTL mapping can aid traditional clinical trait QTL (cQTL) mapping by narrowing the set of candidate genes underlying a given cQTL peak and by identifying expression traits that are causally associated with the clinical traits.
Expression traits detected as significantly correlated with a clinical phenotype may reflect a causal relationship between the traits, either because the expression trait contributes to, or is causal for, the clinical phenotype, or because the expression trait is reactive to, or a marker of, the clinical phenotype. However, correlation may also exist in cases when the two traits are not causally associated. Two traits may appear correlated due to confounding factors such as tight linkage of causal mutations (Schadt et al. ) or may arise independently from a common genetic source. The mouse provides an example of correlations between eumelanin RNA levels and obesity phenotypes induced by an allele that acts independently on these different traits, causing both decreased levels of eumelanin RNA and an obesity phenotype. More generally, a clinical and expression traits for a particular gene may depend on the activity of a second gene in such a manner that conditional on the second gene, the clinical and expression traits are independent.
Correlation data alone cannot indicate which of the possible relationships between gene expression traits and a clinical trait are true. For example, given two expression traits and a clinical trait detected as correlated in a population of interest, there are 112 ways to order the traits with respect to one another. If we consider the traits as nodes in a network, then there are five possible ways the traits (or nodes) can be connected: connected by an undirected edge, connected by a directed edge moving left to right, connected by a directed edge moving right to left, connected by a directed edge moving right to left and a directed edge moving left to right, and not connected by an edge. Since there are three pairs of nodes, there are 5 × 5 × 5 = 125 possible graphs. However, because we start with the assumption that the traits are all correlated with one another, we exclude 12 of the 125 possible graphs in which one node is not connected to either of the other two nodes, in addition to excluding the graph in which none of the nodes are connected, leaving us with 112 possible graphs (Fig. A). The joint trait distribution induced by these different graphs are often statistically indistinguishable from one another (i.e., they are Markov equivalent, so that their distributions are identical), making it nearly impossible in most cases to infer the true relationship. On the other hand, when the two traits are at least partially controlled by the same genetic locus and when more complicated methods of control (e.g., feedback loops) are ignored, the number of relationships between the QTLs and the two traits of interest can be reduced to three basic models illustrated graphically in Fig. B. The dramatic reduction in the number of possible graphs to consider is mainly driven by the fact that changes in DNA drive changes in phenotypes and not vice versa. That is, while it may be possible that changes in RNA or protein lead to changes in DNA at a high enough frequency to detect associations between germ-line transmitted DNA changes and phenotype in segregating populations, it seems extremely unlikely.
It is important to note here that when we use the term causality, it is perhaps meant in a more nonstandard sense than most researchers in the life sciences may be accustomed to. In the molecular biology or biochemistry setting, claiming a causal relationship between, say, two proteins usually means that one protein has been determined experimentally to physically interact with or to induce processes that directly affect another protein and that in turn leads to a phenotypic change of interest. In such instances, an understanding of the causal factors relevant to this activity are known, and careful experimental manipulation of these factors subsequently allows for the identification of genuine causal relationships. However, in the present setting, the term “causal” is used from the standpoint of statistical inference, where statistical associations between changes in DNA, changes in expression (or other molecular phenotypes), and changes in complex phenotypes like disease are examined for patterns of statistical dependency among these variables that allows directionality to be inferred among them, where the directionality then provides the source of causal information (highlighting putative regulatory control as opposed to physical interaction). The graphical models (networks) described here, therefore, are necessarily probabilistic structures that use the available data to infer the correct structure of relationships among genes and between genes and clinical phenotypes (Schadt and Lum ). In a single experiment with one time point measurement, these methods cannot easily model more complex regulatory structures that are known to exist, like negative feedback control. However, the methods can be useful in providing a broad picture of correlation and causative relationships, and while the more complex structures may not be explicitly represented in this setting, they are captured nevertheless given that they represent observed states that are reached as a result of more complicated processes like feedback control.
The classic reductionist view applied to genetics has motivated the identification of single genes associated with disease as one means of getting a foot into disease pathways. However, even in cases where genes are involved in pathways that are well known, it is unclear whether the gene causes disease via the known pathway or whether the gene is involved in other pathways or more complex networks that lead to disease. This was the case with , a recently identified and validated obesity susceptibility gene (Schadt et al. ). The classic view of the signaling pathways involving the superfamily of transforming growth factor β (TGF-β) proteins is that TGF-β acts through receptor serine/threonine kinases to phosporylate regulatory proteins of the Smad family, which then move into the nucleus where they bind DNA to activate specific sets of target genes (Alberts ) (Fig. A). Although the number of biological functions this cascade ultimately impacts is large, the classic pathway is simplistic, involving only a limited number of genes, with little insight provided into the vast network of gene interactions that potentially modulate key players in this pathway.
RNA levels of the type II TGF-β receptor () were recently shown to be very significantly correlated with thousands of other gene expression traits in the liver transcriptional network of a cross between two inbred lines of mice (referred to here as the BXD cross) (Schadt et al. , ) This set of genes associated with was enriched for a broad range of biological functions known to be associated with the classic TGF-β signaling pathway and with metabolic disease traits such as obesity. Furthermore, RNA levels in the BXD cross were also found to be significantly correlated with many obesity-related traits like fat mass, percent body fat, and weight. Taking a view that a complex network of gene interactions underlies obesity phenotypes in the BXD cross, genotypic and gene expression data were systematically integrated to assess whether changes in DNA sequence at a given location in the genome (reflected as genotypes in the cross animals) leading to changes in transcript abundances for a given gene supported an independent, causative, or reactive function of that gene relative to various obesity phenotypes like fat mass (Schadt et al. ). In partitioning the thousands of genes associated with obesity in this way, was one of 40 genes predicted as causal for obesity in the BXD cross. and two other genes selected for validation were all validated as causal for obesity in this study (Schadt et al. ). These data directly demonstrated that and other genes in this signaling pathway are involved in a more general gene network (Schadt et al. , ), so that it is possible that perturbations in these other genes or in itself may drive diseases like obesity by influencing other parts of the network beyond the TGF-β signaling pathway (Fig. B). Therefore, considering single genes in the context of a whole-gene network may provide the necessary context within which to interpret the disease role a given gene may play.
Networks provide a convenient framework for exploring the context within which single genes operate. Networks are simply graphical models comprising nodes and edges. For gene networks associated with biological systems, the nodes in the network typically represent genes, and edges (links) between any two nodes indicate a relationship between the two corresponding genes. For example, an edge between two genes may indicate that the corresponding expression traits are correlated in a given population of interest (Zhu et al. ), that the corresponding proteins interact (Kim et al. ), or that changes in the activity of one gene lead to changes in the activity of the other gene (Schadt et al. ). Interaction or association networks have recently gained more widespread use in the biological community, where networks are formed by considering only pairwise relationships between genes, including protein interaction relationships, coexpression relationships (Gargalovic et al. ; Ghazalpour et al. ), and other straightforward measures that may indicate association between two genes.
Genetic data can aid in the construction of association networks by helping to reduce artifactual correlations between expression traits. Significant artifactual correlations can arise because of correlated noise structures between array-based experiments networks. One way to leverage the eQTL data in this setting is to simply filter out gene-gene correlations in which the expression traits are not at least partially explained by common genetic effects (Lum et al. ). For example, we can connect two genes with an edge in a coexpression network if the value for the Pearson correlation coefficient between the two genes is less than some prespecified threshold, and the two genes had at least one eQTL in common. This can be taken a step further by formally assessing whether two expression traits driven by a common QTL are related in a causal or reactive fashion, filtering out correlations driven by expression traits that are independently driven by common or closely linked QTLs (Doss et al. ; Schadt et al. ).
As has been discussed, multiple traits driven by common QTLs is a central idea that can be leveraged to construct networks. One intuitive way to establish whether two genes share at least one eQTL is to perform single-trait eQTL mapping for each expression trait and then consider eQTLs for each trait overlapping if the corresponding LOD for the eQTLs are above some threshold and if the eQTLs are in close proximity to one another. The significance of the statistic corresponding to the strength of association between two genes in the coexpression networks is then chosen such that the resulting network exhibits the scale-free property (Gargalovic et al. ; Ghazalpour et al. ; Lum et al. ) and the false discovery rate for the gene-gene pairs represented in the network is constrained. Beyond the simple, albeit intuitively appealing, eQTL overlap method, we can formally test whether two overlapping eQTLs represent a single eQTL or closely linked eQTLs by employing a pleiotropy effects test (PET), such as that originally described by Jiang and Zeng () and Zeng et al. (). The formation of gene clusters by simultaneously considering gene-gene and marker-gene correlations also promises to provide a more comprehensive characterization of shared genetic effects (Lee et al. ).
The identification of DNA polymorphisms that associate with diseases like obesity and diabetes can be considered only as the beginning in a long series of steps needed to elucidate disease pathways and to establish the specific role individual genes may play in the process. Diseases like obesity are diseases of the system, potentially involving many different pathways operating in many different tissues and ultimately giving rise to not only different disease subtypes but to different comorbidities of the disease as well. The integration of gene expression (and other molecular profiling data more generally) and genotypic data will be critical if we are ever to understand how genetic and environmental perturbations to a given system lead to complex traits like disease. If common forms of these diseases represent states of a network, then focusing on single-gene perturbations will likely never reveal the most effective ways to treat or prevent disease.
The integration of the diverse sets of molecular data now being generated in population settings is only in its infancy. Many of the methods employed to date toward this end are more heuristic in nature and so will benefit from a more formal treatment. In addition, little to date has been done to integrate expression data from multiple tissues to dissect how modules in one tissue may communicate with modules in another tissue. The types of interactions considered along with eQTL data so far have been restricted to RNA-RNA association data, despite the availability of large-scale DNA-protein and protein-protein interaction data. The predictive power of the types of networks discussed in this review could be enhanced by more systematically integrating protein-protein interactions, protein-DNA interactions, protein-RNA interactions, RNA-RNA interactions, protein state information, methylation state, and interactions with metabolites as these types of data become available. These developments promise to take us beyond the single-gene view of disease and move us closer to the type of systems level view, depicted in Fig. , that may be needed to fully understand the complexity of common human diseases like obesity and diabetes. Of course, further study and experimentation are needed to demonstrate more convincingly that understanding the state of a given molecular network, interactions among molecular networks, and how the states of such networks change in response to different genetic and environmental contexts is tractable enough to take us beyond the reductionist approach, which to date has achieved great success in elucidating the complexity of living systems more generally. |
The tail suspension test (TST) is used to screen for antidepressant activity in pharmacologic studies (Cryan et al. ; Steru et al. , . When a mouse is suspended by its tail, the initial response is to struggle, but this is followed by episodes of immobility, which is taken as an index of its depressive state, paralleling behavioral despair. Duration of immobility is found to be markedly reduced in mice administered antidepressants, showing predictive validity in the TST (Cryan et al. ; El Yacoubi et al. ; Vaugeois et al. ). Robust strain differences in baseline TST have been observed, although the strain rankings have varied (Liu and Gershenfeld ; Ripoll et al. ; Trullas et al. ). Studies such as these strongly point to an underlying genetic basis for immobility and so are potentially useful in investigating genes that are accountable for this phenotype. Baseline TST response and TST response to imipramine, along with measures from the open field test and the light-dark box have been explored across 12 inbred strains (Liu and Gershenfeld ) using factor analysis. It was found that baseline TST and imipramine response loaded independently and may thus be genetically independent, but this interpretation is complicated by the fact that the imipramine measurements were taken on a second TST trial, which they observed has greater immobility than the first trial.
Methods for automation of scoring using strain gauges or video analysis have been presented and validated by comparison with hand coding and detection of strain differences and drug responses (Juszczak et al. ; Steru et al. ). Hand coding of course has an element of subjectivity and there are widely differing criteria in use. Automated methods are not entirely free of interrater subjectivity either because the settings for automated measurement of mobility are selected by the experimenter. Furthermore, the precise phenotypic measure may vary depending on the experimental setup which will affect details of video taping and, thus, analysis. Inconsistently measuring behavioral phenotypes in the TST within and across laboratories may reduce the validity of phenotyping data and, in turn, their value for genetic analyses. Most of these issues are not fully overcome with the use of an automated system where mobile behaviors are still subjectively predetermined; however, they offer the advantage of greater flexibility during analysis that enables us to better optimize and more accurately capture the phenotypes reflected in our behavioral data set. A broader series of data points may be evaluated because many of the settings can be altered during analysis, and so it is possible to look more specifically at a range of values that could be attributed to the immobility phenotype. These features should reduce interrater bias and avoid the need to use arbitrary cutoffs. Optimization criteria, other than correlation with hand-coding results, include magnitude of strain difference or drug responses (Juszczak et al. ). It is likely that different genetic loci affecting TST might differ in their behavioral profile, motivating us to couple scoring optimization with genetic dissection of the trait.
Several studies have investigated the loci that underlie the baseline immobility phenotype. In a cross of the DeFries High and Low open-field selection strains, Turri et al. () detected loci on chromosomes 3, 5, 11, and 19 affecting their TST measure. Yoshikawa et al. () looked at QTLs involved in immobility and single-QTL analysis demonstrated suggestive linkage for immobility on chromosomes 4, 8, and 14, with borderline significant linkage on chromosome 11 in a F population of mice derived from C57BL/6 and C3H/He inbred strains. In a BALB/cJ × A/J cross, Crowley et al. () detected loci on chromosomes 7, 12, and 19. Each of these studies used a different combination of inbred strains so it is not surprising that at least some different polymorphic loci are found in each case. Nonetheless, the populations from these four studies are all ultimately derived from combinations of strains C57BL/6, A, C3H, and BALB/c, and the lack of overlap is nearly complete, with only the chromosome 19 locus possibly coinciding in Turri et al. () and Crowley et al. (). Using a cross between inbred strains NMRI and 129S6, chosen for maximal TST difference, Liu et al. () detected loci on chromosomes 4, 5, 12, and 18 for basal-line TST. The chromosome 4 locus coincides with that from Yoshikawa et al. (). A different, more proximal peak on chromosome 4 is seen with the related tail suspension-induced hyperthermia phenotype.
We performed a detailed analysis of immobility in the TST, which was the final test performed as part of a battery of nine behavioral tests. Taking advantage of naturally occurring behavioral variation in well-established inbred strains and genetic reference populations, the mice selected in this study included males from the BXD recombinant inbred (RI) panel which is derived from a cross between C57BL/6J and DBA2/J. An RI panel is essentially an F cross immortalized by inbreeding. These allow convenient linkage mapping, using existing genotype data and multiple animals of each genotype. In particular, they are useful for investigating complex traits and preliminary mapping of quantitative trait loci (QTLs). Furthermore, direct estimates of (broad-sense) heritability can be made, which is very useful in optimizing trait measures for genetic mapping and genetic correlation, making them a key reagent for integrating diverse phenotypic data, including molecular phenotypes (Bystrykh et al. ; Chesler et al. ; Manly et al. ; Plomin et al. ).
Male C57BL/6J ( = 16), DBA/2J ( = 9), and BXD RI (24 lines, = 204) mice were generated in the Comparative Biology Unit animal facilities at the Institute of Psychiatry using original stocks purchased from The Jackson Laboratory (Bar Harbor, ME, USA). Mice were weaned at 3 weeks of age and transferred at approximately 8 weeks of age to a separate housing facility where they were singly housed and habituated for 2 weeks before undergoing a battery of behavioral tests. Following the habituation period, the mean ( ± standard deviation [SD]) age of all mice was 79.7 ± 15.8 days. Animals were tested in four batches and all efforts were made to minimize the within- and between-batch variability of the RI lines in terms of age and numbers/strain tested. The mean (± SD) of the RI lines reported in the current study was 8.5 ± 4.24 mice per line.
All mice were singly housed in standard cages measuring 30.5 × 13 × 11 cm, with food (Rat & Mouse No. 1 Maintenance Diet, Special Diet Services, Essex, UK) and water available . The housing room was maintained on a reversed 12:12 light cycle with white lights on from 20:00 to 8:00 hours and red light on in the dark cycle, and all behavioral tests were performed between 09:30 and 19:00 hours. Light intensity in the housing room was 400 lux during the lights-on period and less than 2 lux during the dark period. Four red cluster lights (LED cluster red light No. 310-6757; RS Components Northants, UK) of approximate wavelength 705 nm provided minimal red light during the dark phase, allowing experimenters to work with the mice during their dark phase. Ambient temperature in all rooms was maintained at 21 ± 2°C with 45% humidity level. Sawdust and nesting materials in each cage were changed once a week, but never on the day before or the day of testing to minimize the disruptive effect of cage cleaning on behavior. All housing and experimental procedures were performed in accordance with the UK Home Office Animals (Scientific Procedures) Act 1986.
The behavioral tests were conducted in succession as follows: home cage activity, open field, novel object exploration, elevated plus maze, light/dark box, primary screen of SHIRPA, puzzle box, Morris water maze, and tail suspension test (partially described in Galsworthy et al. , ; Mill et al. ). To circumvent the effect of interexperimenter variability, the same experimenter performed TST in all mice. All mice were tested in a randomized order. Two separate trials were performed on each mouse at approximately the same time of day on consecutive days, with a minimum of 24 h between trials. Mice were moved to the behavioral suite adjacent to the housing room immediately before testing. Lighting under test conditions was set to 350 lux.
Using the mobility detection module in EthoVision version 3.1 (Noldus Information Technology bv, Wageningen, The Netherlands; ), which is an automated tracking system, the video recordings collected for each trial of the TST were tracked and analyzed. The main variables for mobility detection in EthoVision are the difference in pixels between current and previous samples detected, an averaging factor that is used for smoothing, and the thresholds assigned for mobility (Noldus Information Technology bv, ).
Strain identity of all animals was verified by genotyping 11 unlinked SNP markers (rs13475902, rs13475988, rs13459051, rs13459052, rs13459060, rs13476554, rs13459069, rs13478483, rs13459109, rs3708840, rs13482131) across seven chromosomes that distinguish the BXD lines.
Batch differences were removed by regression. The strain composition of the batches was not constant, so it is not guaranteed that the (true) batch means are equal. Nonetheless, regression is conservative because it will remove some strain differences when they are partially confounded with batch differences but it is unlikely to create spurious differences. Estimates of genetic effect size were calculated as SS/(SS + SS), where SS are sums of squared deviations calculated using the lm and anova functions from the stats package of the R statistical environment (R Development Core Team ). Pearson’s correlations were calculated using STATA version 9. Dependent-samples test reported were calculated using the STATISTICA analysis tool.
Strain means and variances were calculated across immobile measures for each trial. These were entered into WebQTL, which is a resource for analysis of RI data with databases of genotype, phenotype, and gene expression data (Wang et al. ), for interval mapping and phenotype correlation.
The same data as were uploaded to WebQTL was also genetically mapped using R/qtl (Broman et al. ), with genotype data downloaded from Genenetwork.org, filtered to retain 801 markers with unique strain distribution patterns. The scanone function was used, with default settings (method = ‘em’, model = ‘normal’). The resulting table of LOD scores was plotted using the image function (R graphics package), to produce and Supplementary Fig. 1, and the persp function used to generate Supplementary The mobility module in EthoVision generates data for frequency and duration measures in the analysis profile. Trait values for duration and frequency of immobility demonstrate the variation within and between strains (Fig. ) and transgression in some lines. We investigated the effects of changing the mobility threshold from 12% to 20% and looked at these across a range of running average intervals (1, 5, and 10). Figure illustrates the effect of the threshold and averaging parameters on frequency and duration of immobility in the population as a whole. The mean frequency and the variance in frequency depend strongly on the averaging interval (decreasing as the average interval increases) and less so on the threshold. In contrast, mean duration of immobility depends on the threshold rather than the running average interval, and its variance increases with increasing averaging interval. Similar patterns were apparent in both trials for each quantitative measure.
A fifth of the data set was scored manually concurrently with the automated system to determine how closely the automated scores reflect manually observed behaviors. For duration of immobility, Pearson’s correlations were positive and significant at < 0.001 (Supplementary Table 1). When the threshold of immobility was set at 12%, the correlation values were between 0.58 and 0.78 in trials 1 and 2. Trial 2 measures of immobile duration were generally better correlated with hand coding than those of trial 1. Increasing the threshold to 20% included mobility that would be disregarded by manual scoring, which indeed was reflected in the lower correlations across the two trials with a range of 0.44–0.61. Frequency measures were poorly correlated to the manual scores, probably because it is difficult to consistently record frequency information by hand. The analysis here focuses on duration of immobility (the measure conventionally used), but there may be a completely different phenotype available from the frequency scores.
In performing two trials, we questioned whether trials 1 and 2 elicited a different or a similar behavioral response in the TST. Figure shows that there was a quantitative difference and dependent-samples -test calculations confirmed that this was significant at the < 0.001 level (Supplementary Table 2), with an increase in the mean time spent immobile during the second trial.
Figure illustrates the fraction of variance attributable to genetic differences (strains) over the 12 combinations of threshold and averaging time. This demonstrates the same trend for both trials and in both frequency and duration, which is that the genetic effect size (eta squared) is highest at the lowest averaging interval and 12% threshold. Although the differences observed were modest, duration of immobility was more heritable in trial 1 than in trial 2. The majority of all heritability estimates for trial 1 are over 0.3, whereas only one estimate in trial 2 met this criterion (Table ).
We calculated whole-genome scans for all 12 combinations of scoring parameters. LOD scores are plotted side by side in Fig. for each trial across these measures. The strongest QTL peaks for trial 1 immobility duration are seen on chromosome 4 (peak LOD score = 4.56, rs13477796) and 15 (peak LOD score = 3.41, rs13459176), and contrary to the expectation from heritability estimates, these are both best seen using the 20% mobility threshold. Linkage is not affected much by the averaging interval. For trial 2, the strongest duration QTLs are on chromosomes 11 (LOD = 3.59, rs13481087) and 18 (LOD = 2.19, gnf18.027.000), and there was a pattern of weaker signals that do not overlap with those for trial 1. These indicate that there are different QTL regions influencing behavioral measures across trait scores for the two trials. Frequency of mobility gives quite a different picture, with a distal chromosome 1 locus for trial 1 (LOD = 2.55, rs6202860), and a stronger signal for trial 2 on chromosome 18, corresponding to that for duration (Supplementary Fig. 1).
Comparable results were obtained with interval mapping using WebQTL. Whole-genome LRS score plots for trial 1, immobile duration, are shown in Fig. . Of particular interest were the linkage differences observed within trials between the two thresholds of immobility and across the averaging intervals. The effect of increasing the threshold altered the QTL peak profile. At a threshold of immobility of 12%, the significance of chromosome 15 was highest with a borderline suggestive significant peak on chromosome 4, while at a 20% threshold the chromosome 4 peak was significant with a reduced effect of chromosome 15. Averaging interval does not have as much of an effect as the threshold, but we do see the best LOD scores for both the chromosome 4 and 15 loci at an average interval of 5, contrary to what one would expect from heritabilities (Table ). The QTL profile of trial 2 is completely different, but also shows a large effect of threshold. Supplementary shows the same data in a different graphical form.
The chromosome 4 interval is approximately 20 Mb and contains a few genes that could be potentially interesting: , , and . Trait correlations using hippocampus consortium gene expression data ( RMA) indicated that is positively correlated with our TST scores for immobile duration (0.76, = 4.77e-06). A much smaller QTL region (<13 Mb) on chromosome 15 included the candidate genes , , , and . The most exciting gene whose expression data correlated with our trait scores in trial 2 was (0.72, = 3.09e-05), which is a glial cell transporter that works together with to reduce cytotoxicity of glutamate (Lehre et al. ).
The use of genetic reference populations such as recombinant inbred panels and their potential for accumulating data across time and between laboratories greatly facilitates understanding complex biological systems (Chesler et al. ). The availability of large-scale molecular phenotype data, chiefly from gene expression studies, is the source of much excitement and new advances. Their usefulness is ultimately determined by the link to phenotypes, and identifying informative measures in animal models to characterize them and translating these findings to other species, including human, remain as challenging as ever.
Recombinant inbred lines offer key advantages in understanding complex phenotypes such as behavioral measures. Measurement of the phenotype on multiple replicates of the same genotype is possible, allowing direct estimation of environmental variance and a well-characterized series of genotypes, giving access to genetic variance. The proportion of overall variance attributable to strain (broad sense heritability or genetic effect size) in general will be lower than those seen from human populations, such as from twin studies, because the degree of polymorphism present in recombinant inbred panels is limited. One obvious use of heritability estimates is for optimizing phenotypic measures to extract the maximum genetic information from, for example, a behavioral test. This could be done with any panel of inbred strains, but a recombinant inbred or other genetically informative panel allows the analysis to go a step further and optimize the measure for detection of association with specific loci. This dissection in our study led to the slightly counterintuitive finding that optimizing on overall heritability would have reduced our ability to detect several loci of relatively large effect. We are not testing such a large number of methods that this effect can be attributed to a survey for chance associations (i.e., multiple testing). Rather, it is likely that other genetic effects are present, most likely due to numerous loci of small effect size, that we do not have the statistical power to detect, and these have a different profile. It is clear that the loci that we do detect respond differently to the analysis parameters that we have looked at here.
While it is not the standard procedure to perform two trials of the TST, we wanted to find out if a different response was elicited between trials. Dependent-samples tests confirmed that there was a significant intertrial mean difference and that in the second trial the mean duration of immobility was greater. These are consistent with findings by Liu and Gershenfeld () where similar differences in the duration of immobility were found between the first and second trial in the TST. It is possible that repeated exposure to the test removes novelty and decreases the anxiogenic response to the TST, altering the underlying phenotypic response. Initial exposure to the TST could have exacerbated the anxiety and struggling behaviors, while the second trial may be measuring a learned depressive-like behavioral response or behaviors similar to chronic mild stress. Liu and Gershenfeld () argue on the basis of factor analysis that baseline strain differences in TST and imipramine response are distinct phenotypes. Although their finding was confounded by testing the imipramine response in trial 2 only, our results do support that trial 2 is at least genetically independent of trial 1 in the TST.
Interval mapping of the trait means and variances in WebQTL generated some linkage results that depended on the parameters defined. These were determined according to our experimental setup and video recording of the TST. The effect of significance seen for QTLs on chromosomes 4 and 15 in trial 1 was opposite at an altered threshold and across a range of averaging factors. Interesting QTL results found in trial 1 include an approximately 20-Mb region on chromosome 4 and a region centromeric on chromosome 15 (<13 Mb). A positive correlation was found with our trait values and hippocampal gene expression data for , which is located within the QTL on chromosome 4. This is potentially interesting because has been implicated in promoting cell growth and differentiation and neurite growth and is involved in cell signaling (Pulido et al. ). The peak LRS marker on chromosome 4, rs3708061,is located within an intron of , a gene known to have multiple nonsynonymous polymorphisms between C57BL/6J and DBA2J and has been identified as a quantitative trait gene for strain differences in sensitivity to seizures from withdrawal of alcohol or benzodiazepines and some chemical convulsants, most strongly for those that act with glutaminergic signaling (Fehr et al. ; Shirley et al. ). This might be of particular interest in connection with a gene in the chromosome 15 region, , which belongs to solute carrier family 1, considered to be involved in high-affinity glial transport of glutamate. Few studies have documented the role of in neurobehavioral disorders; however, a recent report has reported its dysregulation in depressed individuals. Choudary et al. () found that and were significantly downregulated in cortical areas of depressed individuals with concurrent upregulation of AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid) kainate receptor genes. The resultant effect of these genes could potentially cause apoptosis, possibly contributing to hippocampal shrinkage seen in depression (Czeh and Lucassen ). Although the trait correlation with proved to be modestly significant, these expression correlation results are quite interesting and could be investigated further along with the finding that gene expression data correlated with our trait scores in trial 2. Furthermore, there could be a possible interacting effect between these two loci that is not detectable from the current sample. The absence of the same QTL effects in trial 2 with very few peaks suggesting linkage could in part be supported by the QTL effects observed from the anxiety measures found by Turri et al. (). When they dissociated QTLs found for measures of anxiety, prior exposure to the test apparatus diminished the effect of chromosome 15 QTLs. Võikar et al. () also revealed reduced emotionality when repeatedly testing mice through a behavioral screen. The chromosome 15 QTL for measures of anxiety in Henderson et al. () is close to but may not coincide with our chromosome 15 locus, which could be linked. Bolivar and Flaherty () reported a QTL peak for intersession habituation on chromosome 15, which is much further distal.
From these results we conclude that for our experimental setup, exploring a range of average intervals at two different thresholds, the best genetic profile was given at a threshold of 20% and an averaging interval of 5. These results are contrary to the expectation that broad sense heritability estimates are an informative way of dissecting the phenotypic variance attributable to genes. Using EthoVision, Juszczak et al. () investigated mobility in the TST at 2.5% and 3% thresholds but used a higher sample rate (12.5 video frames/second) and did not alter the averaging interval across scores. This highlights that the optimal analysis parameters depend on the details of video recording: resolution, lighting, color, and background.
In the present study TST was performed at the end of a battery of behavioral tests, which was designed to dissect and correlate phenotypes from a range of behaviors with overlapping pathways across a panel of BXD strains. It is possible that the genetic profile obtained in our data set would vary if experimentally naïve mice underwent the TST. However, considering the noninvasive nature of the preceding tests, with the most stressful test performed at the end of the battery, we suspect the differences would be marginal. McIlwain et al. () demonstrated that battery tested and naïve mice displayed task-dependent differences but displayed similar levels of anxiety-related behaviors. Mice were housed in individual cages in our study; therefore, results may differ in comparison to group-housed mice results because individual housing effects in these strains have been studied and shown to have altered behavior in comparison to group housing (Võikar et al. ); however, TST was not included in their test battery. Paradoxically, handling in tests prior to the TST may have reduced habituation to the experimental environment (Võikar et al. ).
Our results show that scoring of behaviors in the TST using an automated system such as EthoVision and linking these to genetic analyses are complementary to extending further the genetic underpinnings of immobility.
p |
A
f
u
n
d
a
m
e
n
t
a
l
a
i
m
o
f
b
i
o
l
o
g
y
i
s
n
o
t
o
n
l
y
t
o
i
d
e
n
t
i
f
y
t
h
e
n
o
r
m
a
l
f
u
n
c
t
i
o
n
o
f
g
e
n
e
s
b
u
t
a
l
s
o
t
o
u
n
d
e
r
s
t
a
n
d
t
h
e
i
r
r
o
l
e
i
n
h
u
m
a
n
d
i
s
e
a
s
e
w
h
e
n
m
u
t
a
t
e
d
o
r
o
t
h
e
r
w
i
s
e
a
b
n
o
r
m
a
l
l
y
a
f
f
e
c
t
e
d
.
T
o
t
h
i
s
e
n
d
,
t
h
e
m
o
u
s
e
c
o
n
t
i
n
u
e
s
t
o
b
e
t
h
e
m
o
d
e
l
o
r
g
a
n
i
s
m
o
f
c
h
o
i
c
e
i
n
m
a
n
y
c
a
s
e
s
b
e
c
a
u
s
e
o
f
e
x
t
e
n
s
i
v
e
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
i
t
s
c
o
m
p
l
e
t
e
d
g
e
n
o
m
e
,
t
h
e
a
v
a
i
l
a
b
i
l
i
t
y
o
f
a
n
i
n
c
r
e
a
s
i
n
g
n
u
m
b
e
r
o
f
g
e
n
e
t
i
c
m
a
n
i
p
u
l
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
,
a
n
d
t
h
e
a
b
i
l
i
t
y
t
o
p
e
r
f
o
r
m
p
h
y
s
i
o
l
o
g
i
c
a
n
d
b
e
h
a
v
i
o
r
a
l
t
e
s
t
s
t
h
a
t
c
a
n
b
e
e
x
t
r
a
p
o
l
a
t
e
d
d
i
r
e
c
t
l
y
t
o
h
u
m
a
n
p
h
e
n
o
t
y
p
i
c
t
r
a
i
t
s
.
I
n
a
d
d
i
t
i
o
n
t
o
i
n
d
i
v
i
d
u
a
l
r
e
s
e
a
r
c
h
g
r
o
u
p
s
e
x
a
m
i
n
i
n
g
s
i
n
g
l
e
g
e
n
e
s
,
a
c
o
m
m
i
t
m
e
n
t
h
a
s
b
e
e
n
m
a
d
e
i
n
r
e
c
e
n
t
y
e
a
r
s
t
o
t
h
e
s
y
s
t
e
m
a
t
i
c
g
e
n
e
r
a
t
i
o
n
o
f
m
o
u
s
e
m
u
t
a
n
t
s
o
n
a
l
a
r
g
e
s
c
a
l
e
u
s
i
n
g
v
a
r
i
o
u
s
f
o
r
w
a
r
d
g
e
n
e
t
i
c
s
s
t
r
a
t
e
g
i
e
s
i
n
b
o
t
h
w
h
o
l
e
o
r
g
a
n
i
s
m
s
a
n
d
e
m
b
r
y
o
n
i
c
s
t
e
m
(
E
S
)
c
e
l
l
s
.
D
e
s
p
i
t
e
t
h
e
s
e
c
o
n
s
i
d
e
r
a
b
l
e
e
f
f
o
r
t
s
,
o
v
e
r
7
0
%
o
f
m
a
m
m
a
l
i
a
n
g
e
n
e
s
s
t
i
l
l
d
o
n
o
t
h
a
v
e
a
c
o
r
r
e
s
p
o
n
d
i
n
g
m
u
t
a
n
t
l
i
n
e
;
h
o
w
e
v
e
r
,
t
h
i
s
s
o
-
c
a
l
l
e
d
“
p
h
e
n
o
t
y
p
e
g
a
p
”
i
s
b
e
g
i
n
n
i
n
g
t
o
c
l
o
s
e
a
s
t
h
e
s
e
p
r
o
j
e
c
t
s
c
o
n
t
i
n
u
e
,
e
s
t
a
b
l
i
s
h
i
n
g
a
l
a
r
g
e
a
n
d
v
a
l
u
a
b
l
e
c
a
t
a
l
o
g
o
f
i
n
h
e
r
i
t
e
d
t
r
a
i
t
s
,
t
h
e
i
r
c
a
u
s
a
t
i
v
e
m
u
t
a
t
i
o
n
s
,
a
n
d
i
m
p
o
r
t
a
n
t
l
y
,
m
u
l
t
i
p
l
e
a
l
l
e
l
e
s
r
e
p
r
e
s
e
n
t
i
n
g
e
a
c
h
o
n
e
f
o
r
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
.
One way to study gene function at the level of the whole organism is by examining the consequences of its inactivation. This may be achieved either by “knocking out” the relevant promoter or coding sequences, or by “knocking in” an inactivating point mutation, deletion, or truncation to disrupt the activity of the corresponding product. With a number of strategies available, including gene targeting by homologous, site-specific, or transpositional recombination, gene trapping, and RNA-mediated interference, knockout technology constitutes the most widely used approach to create loss-of-function alleles in model organisms (Chen and Soriano ; Kuznetsov ; Sangiuolo and Novelli ). The exploitation of inducible promoters and tissue-specific recombinase enzymes also allows the deletion of a gene of interest in a particular organ, cell type, and/or stage of development; such conditional and tissue-specific knockouts provide more accurate and finely tuned systems to study gene function than those generated by conventional constitutive technology (Porret et al. ). A combination of these approaches, in addition to the generation of multiple gene knockouts, has revolutionized the study of many fields of fundamental research, most significantly impacting developmental biology with major insights into the physiology of the hematopoietic, immune, skeletal, cardiovascular, and nervous systems (Shastry ; Sheahan et al. ; Ning et al. ).
Equally important, as technical advances in mapping and mutation screening have facilitated the rapid identification of genetic defects related to human disorders, targeted mutagenesis in the mouse has become an invaluable tool to model and gain mechanistic insights into the pathology and progression of these conditions. For example, disruption of the insulin receptor substrate-2 gene has produced a good biochemical model for type-2 diabetes (LeRoith and Gavrilova ). In addition constitutive heterozygous and conditional knockouts deficient for a variety of tumor suppressor genes faithfully recapitulate many of the clinical symptoms present in the corresponding human cancer predisposition syndromes, including and mice for familial breast cancer or mice for basal cell nevus syndrome (Ghebranious and Donehower ; Hakem and Mak ; Pazzaglia ). Other successes of disease modeling in the mouse from which programs of gene therapy have been initiated notably include -deficient mice for cystic fibrosis (Rosenecker et al. ; Snouwaert et al. ). These and many other knockouts have provided ideal preclinical models for diagnostic development, drug discovery, and targeted therapy testing (Walke et al. ; Zambrowicz and Sands ).
Although existing knockouts account for almost half of the known genes in the mouse genome, only 20% have been described in the literature and/or reported in public databases such as the Mouse Knockout & Mutation Database () or the Mouse Genome Database (). With the advent of fully annotated mouse and human genome sequences and only about 15,000 genes remaining to be disrupted, the National Institutes of Health (NIH) has recently launched the Knockout Mouse Project (KOMP), a $52 million cooperative program over five years to generate a comprehensive public resource of systematic knockout mutations of the mouse genome in ES cells by gene targeting () (Austin et al. ). Each year approximately 500 new ES cell lines will be selected by a peer-review process for production of the corresponding knockout mice, reporter tissue expression analysis, and basic phenotyping. Depending on the findings, a number of those will undergo further characterization, including more detailed and specialized phenotyping and tissue profiling. Complementary high-throughput approaches have been applied to ES cell insertional mutagenesis, or gene trapping, and recent technological advances have extended the scope and value of this method to the generation of conditional knockout alleles in addition to targeted mutagenesis (Branda and Dymecki ; Cobellis et al. ). Recently, two major programs have been established, EUCOMM () and norCOMM (), with the aim of generating over 30,000 new conditional gene-trap ES cell lines for analysis. In addition, these projects aim to collate and distribute compatible tissue-specific Cre recombinase lines in parallel and therefore represent a powerful resource for the selection of a desired gene knockout. By centralizing the rapid and efficient production of mouse knockouts and gene-trap lines and making them readily available to the entire scientific community, such large-scale programs will not only save considerable time and money but also will provide the basis for normalization of comparative phenotypic studies.
The seemingly perfect mouse knockout technology, however, also comes with its limitations and pitfalls. In many instances null embryos are not viable due to developmental defects, precluding the functional study of these genes at later stages of development and in the adult. This particularly applies to knockout mice of tumor suppressors which, with few exceptions, all show embryonic lethality with a distinctive pattern of organ malformations (Ghebranious and Donehower ). Although these mice have provided good models for the study of individual genes in embryonic development and the regulation of differentiation, apoptosis, and cell cycle control during organogenesis (Shastry, ), they have not necessarily been useful for the characterization of gene function in tumorigenesis. The development of conditional and tissue-specific gene-deficiency technologies mentioned above, however, has now overcome this restriction.
While mouse knockouts generally do provide very valuable information about the function of a gene , a number of reports have raised legitimate concerns as to their true value as models of human disease (Hochgeschwender and Brennan ; Routtenberg ); knockout mice often fail to recapitulate the expected clinical symptoms, sometimes producing totally unexpected, conflicting, subtle, or absent phenotypes (Elsea and Lucas ). Moreover, although most mouse genes do perform functions identical to their human homologs, the physiologic differences between these species may greatly influence the phenotypic outcome. This may notably account for differences in tissue specificity; prominent examples include - and -deficient mice whose tumor spectrum considerably differs from that seen in patients with Li-Fraumeni and inherited childhood retinoblastoma, respectively (Ghebranious and Donehower ). Another common pitfall of gene targeting is the potential disruption of transcriptional control elements that govern the expression of neighboring genes by the introduction of a selection marker from the targeting vector. This likely explains why very different phenotypes have been obtained from the disruption of the same gene using different vectors. (Gingrich and Hen ; Olson et al. ). A number of reports have also highlighted the fact that the strategies used for gene inactivation are not equivalent. To study the role of phosphoinositide 3-kinase γ (PI3Kγ) in cardiac function, deficient mice were generated either by using a traditional knockout strategy or by knocking in a targeted mutation that causes loss of kinase function. Surprisingly, these mice displayed different phenotypes with PI3Kγ knockout mice showing reduced inflammatory responses and increased cardiac contractility, while the mutant PI3Kγ knockin mouse retained only the immunologic defect. Molecular studies later revealed that PI3Kγ also functions as a scaffolding protein, thus transducing both kinase activity-dependent and -independent signaling pathways (Patrucco et al. ). This demonstrates that different functions of a gene may be revealed depending on the region that is inactivated.
Finally, regardless of the engineering strategy or model organism used, one major risk associated with the complete loss of function of a given gene is the establishment of compensatory mechanisms that may—at least partially—mask the effect of its inactivation. For example, despite the essential role of myoglobin in oxygen transport from erythrocytes to mitochondria, myoglobin knockout mice show normal cardiac function (Garry et al. ). Further studies revealed that more than half of null embryos die and that those surviving have developed adaptative mechanisms to compensate for the defect in oxygen transfer (Meeson et al. ). In that respect, conditional knockout models offer another advantage over their constitutive counterparts as adaptive responses are unlikely to take place shortly after the knockout event. The lack of overt phenotype is especially frequent when the gene of interest belongs to a family of related proteins with some functional redundancy, and in many cases it may be necessary to create animals carrying null alleles of two or more members to obtain informative phenotypes (Kono et al. ).
Knockout technology has become an invaluable experimental tool for assigning gene function and modeling genetic disorders . However, it must be used with caution and some of the examples mentioned above highlight the fundamental problem posed by using genetic deficiency regarding the validity and phenotypic interpretation of knockout models; the studies are often limited to examining the compensatory effects of gene ablation as opposed to changes in the function of the gene of interest. Therefore, to address complex questions regarding gene function and regulation, complementary approaches that alter gene structure in a more subtle way must be used in conjunction with knockout technology, such as the analysis of single-point mutations and their phenotypic consequences. Such strategies may therefore be limited to the modeling of single-gene disorders as opposed to complex traits or mitochondrial or chromosomal disorders; however, approximately 10% of all human genes are currently implicated in monogenic Mendelian diseases, and this number continues to rise (Antonarakis and Beckmann ; Hamosh et al. ). Moreover, at the time of writing over two thirds of the 2300 disease genes in the Human Gene Mutation Database contain point mutations () (Stenson et al. ). In addition, the fact that 75% of disorders in man are inherited in a dominant manner (McKusick ) indicates that both dominant and recessive mouse point mutants are vital to complement targeted knockouts and provide the most clinically relevant disease models.
Before the advent of molecular biology, many important discoveries were derived from mutations that arrived spontaneously in inbred mouse colonies. However, to generate significant numbers of mutant mice more efficiently, a number of large-scale -ethyl--nitrosourea (ENU) mutagenesis programs were instigated ten years ago (Brown and Peters ; Hrabe de Angelis et al. ; Nolan et al. ). Their success led to the establishment of over a dozen independent centers worldwide, each with particular skills and interests and with the aim to standardize procedures and share resources (summarized in Table in Cordes ). Details of these strategies and variations in the breeding schemes have been reviewed extensively elsewhere (Brown and Balling ; Hrabe de Angelis and Balling ; Justice ). Initially, these centers have concentrated on a phenotype-driven approach, in which mutant progeny are screened for abnormalities using a simple yet quantitative assessment of physiology and behavior in combination with more focused phenotyping methods for a specific trait or organ of interest (Keays and Nolan ; Rogers et al. ; Thaung et al. ). Once inheritance of the phenotype is confimed, mutant lines of interest are then analyzed in more detail and the causative mutation is identified by positional cloning.
One advantage of this approach over targeted mutagenesis is it can create a range of mutations: hypomorphic (reduced amount of gene product), hypermorphic (increased amount of gene product), and neomorphic (altered function) alleles, in addition to those that are null (loss of function), facilitating the identification of novel functions for known genes. Such a phenomenon was recently illustrated by our own studies where the identification of a stabilizing gain-of-function point mutation in the mixed-lineage leukemia fusion partner , the cause for the neurodegeneration in the ataxic mouse mutant , revealed a new role for this gene in the central nervous system otherwise unpredictable from the phenotypic presentation of the knockout mouse (Bitoun and Davies ; Bitoun et al. ; Isnard et al. ; Isaacs et al. ; Oliver et al. ).
The random nature of ENU also means that mutiple mutations in the same gene, an allelic series, may occur in independent lines. A combination of such mutants is therefore more likely to provide information related to gene dosage, or the identification of functionally important protein interacting domains, than a classical gene knockout. This may be particularly applicable to the pharmaceutical industry, because typically the mode of action of drugs is to alter the activity of proteins rather than eliminate their function, or to target specific residues of an active site, for example (Russ et al. ). Alternatively, a knockin of the desired mutation might provide a suitable genetic model; however, the time and cost constraints of generating one or more mutant lines using this method is often prohibitive. An engineered mutant trangene might provide a more rapid solution, although studies of transgenic lines are often confounded by factors such as epitope tags, multiple insertions, incorrect promoter artifacts, or position effects that do not influence ENU point mutations that have the advantage of always occurring at the endogneous genomic position. Consequently, an ENU mutation that is not known to be causative in human disease is still likely to provide valuable insights into gene function.
However, the practicalities of large-scale phenotype-driven mouse mutagenesis, such as the limitation and bias of the phenotyping methods, means that many potentially interesting or subtle phenotypes may simply not be detected in a first-pass screen; consequently, mutagenesis centers routinely cryopreserve tissues from each new mutant line for future rederivation and genotyping regardless of the phenotypic data obtained (Glenister and Thornton ). These resources, in combination with recent advances in the rapid detection of mutations by denaturing high-performance liquid chromatography (DHPLC) (Dobson-Stone et al. ) and temperature gradient capillary electrophoresis (TGCE) (Culiat et al. ; Sakuraba et al. ), have made it practical to use a gene-driven approach to mouse mutant detection. Here, a gene of interest can be efficiently screened for mutations by PCR from thousands of individual DNA samples followed by rederivation of the selected lines for further study (Coghill et al. ; Michaud et al. ). This technique is also applicable to ES cells that are amenable to random chemical mutagenesis and PCR screening (Chen et al. ; Munroe et al. ). Although there is no guarantee of any measureable phenotype in the resulting mutant, it has been calculated from pilot studies that 5000 DNA samples is sufficient to identify at least two alleles with 90% confidence (Coghill et al. ; Quwailid et al. ). Such values are based on estimating the proportion of the genome that is protein coding; therefore, assuming no positional bias in the mutagenic action of ENU, the larger the gene, the greater the likelihood of identifying a new mutation (Concepcion et al. ). With the size of these archives increasing and other academic centers such as RIKEN generating similar resources (Sakuraba et al. ), gene-driven screening will play an increasingly important role in the identification of multiple mutant alleles. Another, sometimes overlooked, advantage of such strategies is that all the resulting mutants from a particular ENU screen will be derived from the same genetic background, a vital feature for comparative assessment. It is well known that inbred lines differ considerably in a large number of physiologic and behavioral parameters (Contet et al. ; Kaku et al. ; Solberg et al. ), which may confound attempts to accurately compare spontaneous mutant or knockout lines on different backgrounds (Lalouette et al. ; Runkel et al. ).
#text
The cost of generating a new allelic series is continually decreasing with advances in DNA analysis and the expansion of mutant mouse archives. For example, combining TGCE with over 17,000 mutant genomic DNA samples, Ingenium Pharmaceuticals state they are able to generate at least five new alleles for any given gene and provide live adult mutant mice within three to four months (Augustin et al. ). With this number of novel strains available, preselection of the most biologically relevant mutations will become routine and necessary in many cases (Grosse et al. ). Improvements in ES cell technology have also enhanced the efficiency of gene-driven screens from mutagenized stem cell archives. Recovery of these lines has traditionally relied on introducing ES cells into blastocycts, followed by two rounds of breeding to generate mice homozygous for the desired mutation. Hybrid ES cell lines and tetraploid host embryos can facilitate the identification of phenotypes in the first generation (F), although this method suffers from a number of confounding technical inefficiencies that preclude its use as a high-throughput strategy. The latest advance uses laser-assisted injection of ES cells into eight-cell-stage embryos, generating viable F mice of both sexes that are nearly 100% chimeric. A pilot study demonstrated that the phenotype of previously characterized mutations could be successfully recapitulated in mice recovered from ES cells using this method (Poueymirou et al. ). It has even been possible to identify and successfully rederive new splice variants from a highly pooled archive of 40,000 mutagenized ES cell clones using nested exon-skipping PCR primers (Greber et al. ). Because the time and cost of these high-thoughput technologies is decreasing, the availability of an “off-the-shelf” series of mutations for a given gene is slowly becoming a reality.
It must not be overlooked, however, that each new ENU mutant line harbors many other mutations in addition to the one that may have been identified by gene-driven screening (Hitotsumachi et al. ). The estimate that one functional animo acid change is obtained every 1.82 Mb of coding DNA was determined from a gene-based screen of the Harwell ENU archive (Quwailid et al. ). Consequently it was calculated that there is still a 7% chance that a second confounding mutation is linked to the originally identified loci after ten generations of backcrossing to a wild-type strain (Keays et al. ). Marker-assisted selection (MAS), in which offspring with the smallest amount of donor chromosome linked to the mutation of interest are used for breeding, can be used to circumvent this problem (Visscher et al. ). Although this may detract from the apparent speed and convenience of a gene-driven approach, additional experimental evidence such as a BAC rescue (Keays et al. ) or even a second allelic mutation (Hafezparast et al. ) can provide sufficient supporting evidence that a novel genetic lesion is causative.
As the studies above have illustrated, multiple-mutant alleles frequently provide valuable insight into gene function, including unexpected and serendipitous findings, as a consequence of the random nature of ENU mutagenesis. The power of this technique relies on evolutionary conservation of DNA sequence and physiologic parameters to extrapolate conclusions to human disease states; consequently, the mouse will continue to provide important and clinically relevant phenotypes. Moreover, there is likely to be a move toward mutation screening of non-protein-coding regions such as promoter elements, introns, and noncoding RNAs as more is learned about their role in biology. For example, the modeling of human point mutations in the noncoding RMRP RNA has provided new insight into the role of this component of the ribonucleoprotein complex in cartilage-hair hypoplasia (Hermanns et al. ). There is no doubt, therefore, that large-scale chemical mutagenesis, in combination with other genetic tools such as conventional and conditional knockouts, knockins, and transgenics will continue to play a vital role in the generation of new therapeutic targets. |
xref
#text
MIP superfamily members are typically 28–30 kDa in size and construct an integral membrane pore, characterized topographically by six transmembrane spanning domains, with cytosolic amino and carboxy termini (Fig. ). Structural and amino acid similarities between the first and second half of the protein and comparative analysis of gene structure in paralogous sequences indicate that AQPs likely arose by way of an intragenic duplication that occurred relatively early in evolution (Pao et al. ). Intracellular loop B and extracellular loop E contain highly conserved asparagine-proline-alanine (NPA) motifs which are inserted into the membrane to create the functional water pore, generating what is referred to as the “hourglass model” (Jung et al. ). In some family members, a cysteine residue in the extracellular loop E (Cys 189 in human AQP1) that is situated close to the pore confers functional channel inhibition by mercurials through physical blockage of the pore (Preston et al. ; Zhang et al. ). Recent studies have indicated that the water permeability of a subset of aquaporins can be blocked by quaternary ammonium compounds such as tetraethyl ammonium (TEA) (Detmers et al. ). AQPs assemble as homotetramers in the membrane; however, each monomer is a functional water pore that supports bidirectional, osmotically driven transmembrane water flow. , the direction of water flow through AQPs is determined by the osmotic gradient that exists across the membrane during specific physiologic processes such as absorption or secretion driven by active ion transport.
The NPA motifs are present in nearly all MIP family members, with a few exceptions. Among the 13 known mammalian MIP family members, AQP7, AQP11, and AQP12 encode variant versions of the NPA box in loop B, and AQP7 also has a variant second NPA box in loop E. The proline in the NPA box of loop B of AQP7 is changed to an alanine, thereby changing the first NPA motif to NAA, whereas a serine replaces the alanine in loop E, resulting in an NPS motif (reviewed in Zardoya ). The first NPA box in the B loops of AQP 11 and AQP12 is changed to NPC and NPT, respectively (Gorelick et al. ; Itoh et al. ).
Other primary amino acid sequence differences that exist between members of the MIP family have given rise to two structural classes of proteins whose functions are distinctive. Aquaporins (AQPs) are water-selective members of the MIP family, whereas aquaglyceroporins (GLPs) transport both water and organic compounds such as glycerol, urea (reviewed in Hara-Chikuma and Verkman ), and potentially other small solutes (e.g., NH and NH; Holm et al. ). The determination of pore selectivity for water in the AQP subclass has been examined through a variety of experimental methods, including site-directed mutagenesis, chimeric domain swaps, membrane permeability assays, electron crystallography, X-ray crystallography, and molecular dynamic simulations (reviewed in Gonen and Walz ). X-ray crystallographic analysis of bovine AQP1 from red blood cells at a 2.2-Å resolution identified extracellular and cytosolic pore entry/exit passageways for water molecules that are separated by a central constriction region of the channel (Sui et al. ). The constriction region is formed by the interactions of four amino acids within the pore (His 182, Arg 197, Cys 191, Phe 58), which limits the pore size to a 2.8-Å diameter (Sui et al. ). Three of the amino acids found within the constriction site are conserved in all water-specific MIP members (Arg 197, His 182, Phe 58; Park and Saier ) and contribute to the water selectivity seen in the AQP subclass of the MIP family. In addition, a “selectivity filter,” consisting of six amino acids, was identified. It forces water to make and break hydrogen bonds as molecules pass single file through the pore (Sui et al. ). Using real-time molecular dynamic simulations of water movement through human AQP1, De Groot and Grubmüller () proposed a two-stage filter model in which the NPA motif forms a selectivity-determining region, and a second region termed the aromatic/arginine (ar/R) region functions as a proton filter.
Transport of glycerol has been studied in a few cell types, vertebrate and otherwise. In some cells glycerol crosses the membrane readily, whereas others are quite impermeable (Vom Dahl and Häussinger ). In liver, permeation of glycerol across the membrane (as opposed to its phosphorylation and metabolism) is the rate-limiting step in glycerol utilization (Li and Lin ). Characteristics described for this transport include a combination of simple diffusion and H- or Na-coupled cotransport (Carlsen and Wieth ; Lages and Lucas ; Lucas et al. ). The mechanism is distinct from the glucose transporter and may be phloretin-sensitive (vom Dahl and Häussinger ). These studies may well have incorporated characteristics of multiple transport mechanisms, which remain poorly defined. One mechanism that is common to cells ranging from (Heller et al. ) to insect (Farinas et al. ) to mammalian kidney involves glycerol transport via proteins from the aquaporin family. The “selectivity” of the AQP subclass was determined based on comparisons made with members of the GLP subclass. The best studied member of the GLP subclass is , a glycerol facilitator isolated from that is permeable to glycerol, urea, and glycine, with very low water permeability (Borgnia and Agre ; Maurel et al. ). Although assembles a transmembrane structure that is roughly similar to AQP1, the channel is asymmetric, and the lengths of the extracellular loops and constituents at several amino acid positions within the constriction region and selectivity filter differ significantly from AQP1 (Lu et al. ; Sui et al. ; reviewed in Gonen and Walz ). Five amino acid positions (P1-P5) located in transmembrane helix 3 (P1), extracellular loop E (P2, P3), and transmembrane helix 6 (P4, P5) differ consistently between the AQP and GLP subfamilies in both mammalian and nonmammalian species (Table ; Fig. ) (Froger et al. ; Heymann and Engel ; Lagree et al. ). X-ray crystallographic images in the absence of glycerol confirmed the results of the functional studies showing that does facilitate water transport (Sui et al. ; Tajkhorshid et al. ), although -mediated water permeability is far less than glycerol permeability. Since then the structure of several members of both the AQP and GLP subfamilies in a variety of organisms has been resolved at the atomic level (reviewed in Gonen and Walz ).
Thirteen functionally and phylogenetically distinct mammalian water channels (AQP0–AQP12) have been identified on the basis of sequence homology to AQP1. Evolutionary comparison of mammalian MIP sequences classify AQP0, AQP1, AQP2, AQP4, AQP5, AQP6, and AQP8 as members of the water-selective aquaporin subgroup, whereas AQP3, AQP7, AQP9, and AQP10 are evolutionarily grouped as aquaglyceroporins (Gonen and Walz ; Gorelick et al. ; Zardoya ) (Fig. , Table ). AQP11 and AQP12 are the most distantly related paralogs (Morishita et al. ). They have only approximately 20% homology with the MIP family and may constitute a third functionally distinct evolutionary branch of the MIP superfamily (Gorelick et al. ; Itoh et al. ; Morishita et al. , .
Typically, MIP family members are functionally characterized for osmotic water permeability by expressing the candidate AQP/GLP cRNA in oocytes and assessing cell volume changes upon hypotonic challenge (Preston et al. ). Likewise, solute (i.e., urea/glycerol) permeability can be determined by measuring solute uptake in isotonic solution. In some instances, AQP/GLP function has also been assessed in reconstituted proteoliposomes (Zeidel et al. ) or by expression in yeast (Lagree et al. ). Functionally, AQPs 0–10 support various levels of transmembrane water permeability (Table ). AQP0, AQP6, AQP9, and AQP10 show low water permeability compared with AQP1, AQP2, AQP3, AQP4, AQP5, AQP7, and AQP8. The water permeability of AQP3 and AQP6 is affected by pH. AQP3, AQP7, AQP9, and AQP10 are also permeable to both urea and glycerol, whereas AQP8 has been reported to exhibit urea and ammonia permeability (Saparov et al. ). Interestingly, both AQP7 and AQP9 have been reported to facilitate arsenite uptake (Liu et al. ), and AQP6 functions as an anion channel, with permeability to nitrate and chloride (Ikeda et al. ; Yasui et al. ). The membrane transport properties of AQP11 and AQP12 are currently unknown. Evaluation of AQP11 function in oocytes failed to identify any water, urea, glycerol, or ion permeability under various pH conditions (Gorelick et al. ). The transport properties of AQP12 remain unstudied because of the inability to obtain adequate plasma membrane expression in oocytes (Itoh et al. ).
#text
Some of what is known of the physiologic function of AQPs has been learned by examining the clinical manifestation and associated pathology of AQP deficiency in human disease. Human disorders whose pathogeneses are associated with defects in water channel proteins include inherited cataracts (AQP0; Berry et al. ; Francis et al. ) and nephrogenic diabetes insipidus (AQP2; Deen et al. ; reviewed in Knoers and Deen ), and water channel dysfunctions have been implicated in the etiology of Sjogren’s syndrome (AQP5; Beroukas et al. ; Steinfeld et al. ; Tsubota et al. ). Individuals lacking functional AQP1 have a decrease in pulmonary vascular permeability (King et al. ) and fail to concentrate urine maximally when dehydrated (King et al. ). The list of AQP involvement in human disease is likely to increase as an understanding of the complexity of AQP function becomes better realized.
Targeted disruptions of individual water channel proteins in knockout mouse models have been useful in further elucidating the role of AQP/GLPs in whole-animal physiology. Table summarizes the phenotypes reported to date for the AQP-deficient mice currently available. The information gleaned from the knockout strains implicates abnormal AQP function/regulation as a potential contributor to other urine-concentrating defects (AQP3; Ma et al. ) and to human diseases such as disorders of the skin (AQP3; Ma et al. ), impaired wound healing (AQP3; Hara et al. ), hearing loss (AQP4; Li and Verkman ), salivary gland secretory defects (Ma et al. ; Krane et al. ), impaired sweat gland function (Nejsum et al. ), asthma (AQP5; Krane et al. ), diabetes and insulin resistance (AQP7; Hibuse et al. ), obesity (AQP7; Hara-Chikuma et al. ) and polycystic kidney disease (AQP11; Morishita et al. ). As such associations emerge for human pathologies, designed pharmacologic inhibition of AQP function may be of specific clinical utility, such as for preventing cellular migration during metastatic tumor progression (AQP1; Saadoun et al. ) or in response to brain edema or ischemic injury (AQP4; Manley et al. ).
Aquaporin proteins are found ubiquitously among the kingdoms of living things. Many of these organisms experience extremes of thermal and osmotic stress far beyond those tolerated by mammals. One such circumstance is the possibility of substantial cellular dehydration elicited by freezing; organisms ranging from bacteria through certain vertebrates tolerate subfreezing temperatures, and they do so using a combination of water and solute transport mechanisms.
Recently, the importance of AQP/GLP in the process of glycerol-facilitated cryopreservation has been shown under natural and experimental conditions. Glycerol is an organic solute commonly used in biomedicine as a cryoprotectant to enable bacterial, fungal, and embryonic cells to freeze at ultralow temperatures without compromised viability. Enhanced AQP/GLP expression correlates with improved freeze tolerance in baker’s yeast (Tanghe et al. ) and sperm (Dibas et al. ), and artificial AQP/GLP expression improves viability following cryopreservation of fish embryos (Hagedorn et al. ) and mouse oocytes (Edashige et al. ), suggesting that facilitated glycerol transport through AQP/GLP may participate in the physiology of freeze tolerance in animals. Clinical cryopreservation is currently most successful with small or single-celled tissues. Therefore, insights into how a multicellular organism survives freezing could yield important clues to the cryopreservation of larger tissues and organs. For example, modeling studies have suggested that cryopreservation of whole mammalian kidneys, which would entail perfusion of cryopreservative solution such as glycerol through the vasculature, would succeed best if tissue permeability to the cryopreservative agent were high, thereby minimizing osmotically induced changes in cellular and extracellular volumes (Lachenbruch et al. ). Thus, the comparative analysis of AQPs/GLPs in amphibians, a subset of which undergo a physiologic process of cryopreservation and freeze tolerance (sometimes involving glycerol accumulation), may serve as a model for testing such ideas.
Renal conservation of cryoprotective solutes may be critical to freeze-tolerant anurans. Repeated cycles of freezing and thawing deplete glycogen stores in anuran liver (Lee and Costanzo ), quite likely because glucose is lost in the urine: The renal tubules have a limited capacity for glucose reabsorption, and this may be overwhelmed at high glucose concentrations occurring after freezing (Layne et al. ). Wood frogs may compensate for this loss by cutaneous uptake of excreted glucose.
In wood frogs, glycogenesis is initiated promptly upon thawing, thereby minimizing the duration of high plasma glucose and so its potential urinary loss. In contrast, gray treefrogs retain high plasma glycerol concentrations for weeks, before and after freezing. How do they avoid urinary loss of this solute? Presumably glycerol is filtered, albeit at a reduced rate, in cold-acclimated frogs, which have reduced rates of glomerular filtration (Zimmerman et al. ). Thus, reabsorption of filtered glycerol should be at a premium.
In other circumstances when glomerular filtration rate (GFR) is reduced in amphibians, plasma arginine vasotocin (AVT), the amphibian antidiuretic hormone, is elevated (Nouwen and Kuhn ; Rosenbloom and Fisher ). AVT may act on the renal vasculature to reduce GFR (Pang ), on the bladder to increase water permeability, and on the renal tubules (Uchiyama ). The physiologic response of the latter structures to AVT is not known, but one response of vertebrate kidneys to ADH is upregulation and membrane insertion of aquaporins, which could mediate reabsorption of water and/or glycerol.
Historically, amphibians have played a critical role in the conceptualization of and suggestion for the existence of water channels long before the first water channel was cloned. Indeed, even before aquaporins were understood as such, careful study of toad urinary bladder suggested the “shuttle hypothesis,” based on visualization of “particle aggregates” that appeared to be inserted into and retrieved from the apical membrane under conditions of changing water reabsorption (e.g., Wade ; Wade et al. ). Additional evidence has confirmed a role for AQPs and GLPs in amphibian osmoregulation. Aquaporins have been identified in amphibian skin, bladder, fat body, and elsewhere (Ma et al. ; Virkki et al. ; Zimmerman et al. ), and several proteins of the aquaporin family have been sequenced from anurans. Phylogenetic analysis of 17 anuran AQP mRNA sequences deposited in public databases has revealed six classes of anuran AQPs, two of which are distinct to anurans (reviewed in Suzuki et al. ). “FA-CHIP” in (Abrami et al. ), “AQP-t1” in (Ma et al. ), AQP-h1 in (Hasegawa et al. ), and HC-1 in (Zimmerman et al. ) resemble each other in both sequence and wide tissue distribution patterns. These proteins are also similar to mammalian AQP1 (76%–98% sequence identity), and expression cloning has confirmed that AQP-t1, AQP-h1, and HC-1 function as water but not as glycerol channels (Hasegawa et al. ; Ma et al. ; Zimmerman et al. ) (Table ). Temperature-sensitive regulation of HC-1 expression was seen in brain, kidney, and liver; frogs acclimated to cold conditions (4°C) had higher HC-1 mRNA expression in the liver than did warm-acclimated frogs, whereas in brain and kidney warm-acclimated frogs expressed higher levels of HC-1 (Zimmerman et al. ) (Fig. ). A second AQP, HC-2, has recently been isolated from urinary bladder cDNA from and shows strong amino conservation compared with mammalian AQP2 (69% identity, 85% similarity; Zimmerman et al. ). Like hAQP2, HC-2 was functionally determined to be within the AQP subclass, and it supported osmotically driven water transport (Zimmerman et al. ) (Table ). HC-2 mRNA was detected primarily in organs of osmoregulation (skin, bladder, and kidney; Zimmerman et al. ). Interestingly, like HC-1, HC-2 expression varied depending on thermal conditions, i.e., hydrated frogs that were acclimated to cold conditions (4°C) had high levels of HC-2 expression in skin, whereas no HC-2 expression was observed from the ventral skin of hydrated warm-acclimated frogs (Zimmerman et al. ).
To date, four anuran sequences similar to mammalian AQP3 have been identified. They include AQP3 from (Schreiber et al ), AQP from (unpublished; GenBank Accession number CR855446), AQP-h3BL from (Akabane et al. ), and HC-3 from (Zimmerman et al. ). HC-3 from shows 82% identity and 94% amino acid similarity with mammalian AQP3, and functionally it performs as a GLP, with low water permeability and high glycerol permeability (Zimmerman et al. ) (Table ). HC-3 mRNA exhibited both tissue-specific and thermal-selective patterns of expression. Of special note, tissue glycerol concentrations increased in the liver and skeletal muscle in cold-acclimated frogs compared with warm-acclimated frogs (Fig. ). The increase in glycerol concentration in these tissues corresponds well with an increase in HC-3 mRNA abundance in muscle, liver, and bladder in cold-acclimated frogs (Zimmerman et al. ). Studies of mammalian GLPs are just beginning to elucidate potential physiologic roles for their facilitation of glycerol transport. These roles include glycerol export from adipocytes (Hara-Chikuma et al. ) and a contribution to pliability in skin (Ma et al. ). Amphibians that naturally accumulate glycerol represent a natural model for studying the roles and regulation of glycerol-transporting aquaporins.
Two aquaporins from (AQP-h2 and AQP-h3) have been sequenced (Hasegawa et al. ; Tanii et al. ) that have high homology to each other and to AQP-t2 and AQP-t3 from Suzuki et al. () have suggested that these four genes form an anuran-specific, phylogenetically distinct MIP subclass (type AQPa2). AQP-h2 and AQP-h3 are both expressed in ventral skin, whereas AQP-h2 is also expressed in the urinary bladder. Expression of both is upregulated by AVT (Hasegawa et al. ). Coexpression of the AVT receptor, AQP-h2, and AQP-h3 during metamorphosis, when the animals are undergoing a transition from aquatic to terrestrial environment, suggests a role for AVT-regulated AQPs in this process (Hasegawa et al. ). A second anuran-specific phylogenetic class, type-a1, has been assigned for a novel aquaporin identified from oocytes of ; that protein exhibits unique mercury sensitivity and less than 50% amino acid identity to the most closely related mammalian aquaporins (Virkki et al. ).
u
d
i
e
s
o
f
m
a
m
m
a
l
i
a
n
a
q
u
a
p
o
r
i
n
s
h
a
v
e
r
e
v
e
a
l
e
d
m
a
n
y
d
e
t
a
i
l
s
o
f
s
t
r
u
c
t
u
r
e
a
n
d
f
u
n
c
t
i
o
n
a
n
d
a
r
e
b
e
g
i
n
n
i
n
g
t
o
y
i
e
l
d
i
n
s
i
g
h
t
s
i
n
t
o
p
a
t
h
o
g
e
n
i
c
m
e
c
h
a
n
i
s
m
s
.
N
e
v
e
r
t
h
e
l
e
s
s
,
m
a
m
m
a
l
i
a
n
a
q
u
a
p
o
r
i
n
s
c
o
m
p
r
i
s
e
a
r
e
l
a
t
i
v
e
l
y
l
i
m
i
t
e
d
s
u
b
s
e
t
o
f
p
r
o
t
e
i
n
s
f
r
o
m
t
h
i
s
l
a
r
g
e
c
l
a
s
s
o
f
m
o
l
e
c
u
l
e
s
.
P
r
o
t
e
i
n
s
f
r
o
m
t
h
e
M
I
P
f
a
m
i
l
y
a
r
e
p
r
e
s
e
n
t
i
n
e
v
e
r
y
s
o
r
t
o
f
o
r
g
a
n
i
s
m
,
f
r
o
m
b
a
c
t
e
r
i
a
t
h
r
o
u
g
h
f
u
n
g
i
,
p
l
a
n
t
s
,
a
n
d
a
n
i
m
a
l
s
.
I
t
i
s
l
i
k
e
l
y
t
h
a
t
n
o
v
e
l
i
n
s
i
g
h
t
s
i
n
t
o
s
t
r
u
c
t
u
r
e
-
f
u
n
c
t
i
o
n
r
e
l
a
t
i
o
n
s
h
i
p
s
,
i
n
t
o
m
e
c
h
a
n
i
s
m
s
o
f
r
e
g
u
l
a
t
i
o
n
,
a
n
d
i
n
t
o
p
h
y
s
i
o
l
o
g
i
c
r
o
l
e
s
w
i
l
l
d
e
r
i
v
e
f
r
o
m
t
h
i
s
d
i
v
e
r
s
i
t
y
o
f
o
r
g
a
n
i
s
m
s
.
A
m
p
h
i
b
i
a
n
s
p
r
e
s
e
n
t
p
a
r
t
i
c
u
l
a
r
l
y
a
t
t
r
a
c
t
i
v
e
m
o
d
e
l
s
f
o
r
s
t
u
d
i
e
s
o
f
a
q
u
a
p
o
r
i
n
f
u
n
c
t
i
o
n
i
n
v
e
r
t
e
b
r
a
t
e
o
s
m
o
r
e
g
u
l
a
t
i
o
n
a
n
d
t
h
e
r
m
o
r
e
g
u
l
a
t
i
o
n
;
t
r
a
n
s
i
t
i
o
n
s
f
r
o
m
w
a
t
e
r
t
o
l
a
n
d
a
n
d
t
o
l
e
r
a
n
c
e
o
f
t
i
s
s
u
e
f
r
e
e
z
i
n
g
a
r
e
t
h
e
e
p
i
t
o
m
e
s
o
f
c
o
m
b
i
n
e
d
o
s
m
o
t
i
c
a
n
d
t
h
e
r
m
a
l
d
e
m
a
n
d
s
.
T
h
u
s
,
j
u
s
t
a
s
s
t
u
d
i
e
s
o
f
a
m
p
h
i
b
i
a
n
s
c
o
n
t
r
i
b
u
t
e
d
t
o
t
h
e
o
r
i
g
i
n
a
l
e
l
u
c
i
d
a
t
i
o
n
o
f
a
q
u
a
p
o
r
i
n
f
u
n
c
t
i
o
n
,
b
e
f
o
r
e
w
e
k
n
e
w
o
f
a
q
u
a
p
o
r
i
n
s
p
e
r
s
e
,
w
e
s
u
g
g
e
s
t
t
h
a
t
s
u
c
h
s
t
u
d
i
e
s
w
i
l
l
c
o
n
t
i
n
u
e
t
o
c
o
n
t
r
i
b
u
t
e
t
o
a
f
u
n
d
a
m
e
n
t
a
l
u
n
d
e
r
s
t
a
n
d
i
n
g
o
f
t
h
e
s
e
u
b
i
q
u
i
t
o
u
s
p
r
o
t
e
i
n
s
. |
italic
#text
xref
#text
xref
italic
#text
italic
xref
#text
Disorders of neuronal migration are not limited to mice. In humans abnormalities of neuronal migration cause a range of diseases, most notably lissencephaly. Classic lissencephaly, a disorder of both tangential migration and radial migration, is characterized by a cortex that has only four layers and a brain that appears smooth with an absence of gyri and sulci (Dobyns and Truwit ) (Fig. ). Genetic studies have resulted in the identification of several genes that are responsible for abnormal cortical migration in humans.
In two landmark papers, des Portes et al. () and Gleeson et al. () reported that mutations in the X-linked gene doublecortin cause lissencephaly in males and a syndrome known as double-cortex in females. While initially thought to code for a putative signaling protein, it has since been demonstrated that DCX, a protein that is highly expressed in migrating and differentiating neurons, plays a key role in the stabilization of microtubules (Francis et al. ; Gleeson et al. ). Indeed, the mutations in DCX that cause lissencephaly have been shown to cluster in DCX’s two tubulin-binding domains (Sapir et al. ; Taylor et al. ). The creation of transgenic mice lacking have further demonstrated the importance of this gene for neuronal migration. knockout mice have a fractured pyramidal cell layer in the hippocampus and exhibit defects in tangential migration from the subventricular zone to the olfactory bulbs via the rostral migratory stream (Corbo et al. ; Koizumi et al. ). It has also been shown that migrating neuron populations derived from the medial ganglionic eminence in knockout mice show abnormalities in migratory dynamics (Kappeler et al. ).
Lissencephaly is also known to result from hemizygous deletions in a gene on chromosome 17, known as or (Reiner et al. ). contains WD40 domains and is highly expressed in Cajal-Retzius cells and in the ventricular epithelium in the developing human cortex (Clark et al. ). Initially discovered because it catalyzes the inactivation of platelet-activating factor (Hattori et al. ), has been shown to influence microtubule function. Cellular studies have demonstrated that overexpression of increases retrograde movement of cytoplasmic dynein leading to the accumulation of microtubules (Smith et al. ). There is also evidence of an interaction between the reelin signaling pathway and LIS1. It has been shown that phosphorylated DAB1 binds to LIS1 and that compound mutant mice with mutations in both reelin and exhibit enhanced layering defects in the cortex and hippocampus (Assadi et al. ).
The importance of microtubules in neuronal migration is further emphasized by our recent discovery that mutations in α-tubulin cause abnormalities in neuronal migration (Keays et al. ). Microtubules are formed by the polymerization of heterodimers consisting of α- and β-tubulin. We found that an abnormal radial neuronal migration phenotype in the ENU-induced mouse mutant was attributable to a substitution in the GTP binding site of α-tubulin (TUBA1) that affected heterodimerization with β-tubulin. Consequences of impaired neuronal migration in this mouse were notable defects in hippocampal pyramidal cell lamination (Fig. ) and wavelike perturbations in layers II/III and IV of the cortex. The mouse mutant showed striking similarity to the and knockout mice (Corbo et al. ; Hirotsune et al. ), leading us to speculate that mutations in the human homolog of might also cause lissencephaly. This was found to be the case, and to date approximately ten mutations have been found that result in cortical anomalies in humans (K. Poirier et al., personal communication).
So far we can see that the study of human and mouse mutants has unraveled a pathway that involves extracellular guidance cues, intracellular signaling molecules, and a number of proteins that are associated with cytoskeleton stabilization and modulation (Fig. ). How do these mutations operate to affect how a neuron actually moves? Insight into the mechanism was to come from an unlikely source: .
is a filamentous soil fungus. Beginning its existence as a single spore, it undergoes several rounds of nuclear division (resulting in about 4 nuclei), followed by the formation of a projection known as a germ tube. Nuclei then migrate along the germ tube, with each nucleus moving a different distance so the nuclei are separated equally, prior to cellular division (Xiang and Morris ) (Fig. ). Polarized growth of the germ tube then continues, creating an extending tip called a hyphae that undergoes branching and growth. The ability to view this process with a light microscope established as a model organism to study the biology of nuclear migration.
Pioneering experiments with this species were undertaken by Morris. They employed UV radiation to identify a host of nuclear migration mutants, labeling them nuclear distribution (nud) mutants (Morris ). These mutants showed normal germination and nuclear division, but the nuclei failed to migrate into the germ tube at restrictive temperatures. The mutants include , , , , and . They were to be informative in the study of neuronal migration.
Comparative studies showed that NudF, a WD-repeat protein, is 42% identical in amino acid sequence to (Xiang et al. ), the same gene that causes lissencephaly in humans. It was this observation that gave rise to the hypothesis that nuclear migration is required for neuronal migration (Xiang et al. ). Other genes also have mammalian homologs. The homolog of is cytoplasmic dynein heavy chain and that of is cytoplasmic dynein light chain (Xiang et al. ). The dyneins are microtubule-dependent motor proteins that are involved in the motility of a wide variety of organelles. It has been demonstrated that in vertebrates LIS1 colocalizes and interacts with cytoplasmic dynein heavy chain, and both are highly expressed in postmitotic migrating neurons in the cortex (Niethammer et al. ; Smith et al. ). Mice with ENU-induced mutations in cytoplasmic dynein heavy chain are a model for motor neuron degeneration but also exhibit defects in the migration of facial motor neurons; their cell bodies fail to migrate to their final destination in the hindbrain (Hafezparast et al. ). Moreover, disruption of cytoplasmic dynein results in impaired motility in a cellular assay for neuronal migration (Shu et al. ; Tanaka et al. ). The dyneins and their role as microtubule motors are clearly important for nuclear and neuronal migration.
xref
#text
italic
#text |
All vertebrates have pituitary glands composed of specialized hormone-producing cells (Matsumoto and Ishii ). The individual hormones are evolutionarily conserved, although their function varies across the classes of . This conservation suggests that genetic regulation of pituitary function may be conserved.
In humans, growth insufficiency resulting from pituitary hormone deficiency is not infrequent, occurring in approximately 1 in 4000 live births (Procter et al. ; Vimpani et al. ). Growth hormone (GH) insufficiency is the most common type of dwarfism and usually results from mutations in the GH gene cluster (Braga et al. ; Mullis et al. ). Multiple pituitary hormone deficiencies (MPHD) result from mutations in transcription factors important for the normal development and function of the pituitary gland, including (), (), , , , and (Bhangoo et al. ; Cogan et al. ; Laumonnier et al. ; Machinis et al. ; Mendonca et al. ; Netchine et al. ; Pfäffle et al. ; Radovick et al. ; Tajima et al. ; Wu et al. ). The first transcription factor to be linked to MPHD was PIT1 (Tatsumi et al. ). Patients with mutations in generally have deficiencies in GH, prolactin (PRL), and thyroid-stimulating hormone (TSH) as well as profound pituitary hypoplasia (Cohen et al. ). Mutations in are a common genetic cause of familial MPHD. Patients with mutations exhibit progressive hormone loss with varying age of onset and severity (Bottner et al. ; Fluck et al. ). Most common are deficiencies in PRL, GH, and TSH as well as follicle-stimulating hormone (FSH) and luteinizing hormone (LH) (Agarwal et al. ; Deladoey et al. ). Progressive adrenocorticotrophic hormone (ACTH) loss presents as late as the third decade of life (Bottner et al. ). Some MPHD cases, however, cannot be traced to mutations in the protein-coding or intron-exon splice sites in the DNA sequence of known genes important for pituitary development.
The mechanism of action has been studied extensively in two mouse models, the Ames dwarf () and the (Nasonkin et al. ; Ward et al. ). In mice, is expressed throughout the developing Rathke’s pouch in a dorsal to ventral expression gradient from about embryonic day 10 (e10) until about e16 (Sornson et al. ). transcripts are present in the adult pituitaries of human and pig, although the levels were not quantified relative to the embryonic pituitary (Nakamura et al. , ; Skelly et al. ; Sloop et al. ; Usui et al. ). A number of downstream targets have been identified, including , and (Brinkmeier et al. ; Gage et al. ; Raetzman et al. ).
It is of great interest to identify the transcriptional regulators of because it is pituitary-specific, unlike many of the other key regulators of pituitary development: , , , , and . Each of these genes is expressed prior to and is a candidate for transcriptional regulation of . expression is activated in and single mutants, but this may be due to the ability of and to compensate for each other (Suh et al. ; Szeto et al. ). expression also appears to be initiated normally in and deficient mice (Dasen et al. ; Raetzman et al. ). Thus, the spatial and temporal regulation of expression is not fully explained by these genes, suggesting that additional factors may be involved.
The regulation of gene expression involves the cooperation of a variety of transcription factors in a tissue-, temporal-, and/or spatial-specific fashion that interact with -acting regulatory elements in DNA sequences (Kleinjan and van Heyningen ). The identification of these elements can be difficult because they may be located at a considerable distance from the gene or even within the introns of neighboring genes (Bagheri-Fam et al. ; Lang et al. ; Lettice et al. ). The identification of these elements is facilitated by comparative genomics in the form of cross-species DNA comparisons. The alignment of the DNA sequences of orthologous genes from different species, both closely and distantly related, can reveal potential conserved regulatory elements that can then be analyzed (Boffelli et al. ; Nobrega and Pennacchio ; Pennacchio and Rubin ). Transcription factors involved in development, like , are often conserved among vertebrates, and their regulatory sequences are also likely to be conserved (Plessy et al. ).
In this study we report the sequence of the gene from several mammals and utilize cross-species PROP1 protein sequence comparison to verify the conservation of the functional domains of the protein and use genomic sequence comparison to identify putative transcriptional regulatory elements in the noncoding regions of the gene. This analysis revealed the presence of three conserved noncoding elements within and near the gene. Each of them exhibited orientation-dependent enhancer activity in tissue culture, and an element in intron 1 conferred tissue-specific and unique spatial expression in transgenic mice. These studies establish a functional role for the intronic element in gene regulation. Finally, using bacterial artificial chromosome (BAC) transgene rescue of the mutant phenotype, we demonstrate that all of the sequences necessary for functional expression of are located within the BAC.
A BAC clone of approximately 200 kb and containing the pig () gene was identified, and DNA was isolated from it. A shotgun sequencing library was prepared from that BAC by using SeqWright (Houston, TX), and the resulting subclones were sequenced at the University of Michigan Sequencing Core using Applied Biosystems (Foster City, CA) sequencers (Model 3700) and BigDye V1.1 terminator chemistry, according to standard manufacturer’s protocols. Sequence assembly of the shotgun sequence data was performed using phred and phrap (Ewing and Green ), and consed (Gordon et al. ), resulting in 6X draft sequence coverage. Limited finishing was performed based on the “autofinish” option in consed to close some gaps, especially those in the vicinity of the gene. A total of ten draft contigs of 1 kb or greater were obtained in the final assembly, accounting for 196 kb, or an estimated 98% of the original BAC. The sequences have a phred Q-score of 20 or higher, with the majority being Q40 or better. The BAC sequences were submitted to GenBank (NCBI; accession number EF590118). One contig of approximately 26.3 kb contained the gene, including 13.8 kb of 5′ flanking sequence, 3.7 kb encoding the gene, and 8.7 kb of 3′ flanking sequence. A set of lemur BACs containing the gene were isolated from a library derived from a cell line of the ring-tailed lemur (; AG07100C, Coriell Cell Repositories, Camden, NJ) using a human exon 3 probe. One lemur BAC, LBNL-2 102B17, was sequenced from ends of 3 kb subclones to approximately tenfold coverage using BigDye terminators (Applied Biosystems) and assembled into ordered and oriented contigs with the Phred-Phrap-Consed suite (Ewing and Green ; Gordon et al. ). The assembled BAC sequence was submitted to GenBank with accession number AC162436. gene sequence from gorilla (), Black and Red Howler (), Brown Capuchin (), and Gelada Baboon () was amplified from genomic DNA (gift from Dr Deborah Gumucio, University of Michigan, Ann Arbor, MI) by using the forward primer (5′-CCTGCTCCCAGGAGGGGATT-3′) that corresponded to the human 5′ flanking sequence that was highly conserved between mouse and human, and as the reverse primer (5′-AGGCTGGGGATCACCTTGGTG-3′) that corresponded to the 3′ UTR of the human gene. cDNA sequence was determined by using the high conservation between primate exon splice acceptor/donor sites to assemble the cDNA sequence from the gene sequence. The protein sequence was then translated from the cDNA as described above. The gene sequences were deposited in GenBank with accession numbers DQ177426 for gorilla, DQ177425 for capuchin monkey, DQ177427 for howler monkey, and DQ177424 for baboon.
The 20-kb human was obtained from the human chromosome 5 contig sequences from GenBank accession numbers NT086684 and NT023133. The 20-kb mouse () sequence was obtained from the mouse chromosome 11 contig sequences from GenBank accession number NT096135. The genomic sequence for the chimpanzee () was obtained via the Berkley Genome Pipeline from the genome VISTA analysis program [ (Couronne et al. )]. The cDNA and protein sequences were determined as described above. The genomic sequence for the rat and partial genomic sequence for the fugu () and zebrafish () were obtained by searching the UCSC genome browser [ (Kent )] for the closest matches to . The partial protein sequences for the fugu and zebrafish were determined by the translation of the partial gene sequence in all three frames to identify the PROP1 homedomain and transactivation domain sequence. The rat () cDNA sequence was obtained from GenBank (accession number NM153627) and translated as described above. genomic sequence was obtained for the dog (; AF126157) and sheep (; AY533708) and PROP1 protein sequence for human (NP006252), pig (NP001001263), cow (; NP777103), sheep (NP001009767), mouse (P97458), dog (NP001018643), and partial protein sequence for the chicken (; AB037110) was obtained from the NCBI website.
ClustalW alignment for protein and DNA sequences were done with the LASERGENE Navigator Meg align sequence alignment program (DNASTAR Inc., Madison, WI). Mouse, rat, human, and chimp chromosome comparisons were done using the University of California, Santa Cruz (UCSC) genome browser [ (Kent et al. )]. The pig BAC contigs were compared to the mouse genome using the genome VISTA program [ (Couronne et al. )]. The 20-kb genomic sequences for the human, lemur, pig, and mouse were analyzed with the mVISTA [ (Bray et al. )] comparative genomics program to determine the identity of the conserved noncoding elements.
The plasmid was constructed for the targeted knockout of the gene (Nasonkin et al. ). The plasmid was constructed by digesting the plasmid with I and subcloning the region containing 3 kb of the 5′ flanking region, the coding region, and mouse protamine 1 splice and polyadenylation regions into the I site of pBluescript SK+ (Stratagene, La Jolla, CA). A 9.5-kb genomic clone (Nasonkin et al. ), which was generated from a P1 clone containing the entire gene, was used as a template to amplify the region with a series of primers that engineered flanking sites (5′-ATAACTTCGTATAGCATACATTATACGAAGTTAT-3′). The floxed fragment was subcloned into pGEM-T Easy (Promega, Madison, WI). This construct was digested with I/I and ligated into the I/I sites of the pBluescript SK+ vector that contained the 3-kb 5′ flanking sequence with the reporter.
The plasmid was made by creating a chimera of the intron 1 in which the region was deleted. First, the intron sequence for the 5′ flank of was amplified from the 9.5-kb genomic clone (Nasonkin et al. ) by using primers (5′-GGTTTGGGTGGCTAGCCATGGAA-3′ and 5′-TTCCCAAGCACCTCCTTCATATCCCACCCCCCAACTAAGCACCC-3′) that allowed this fragment to be annealed to the 3′ flanking sequence that was also amplified from the 9.5-kb plasmid with the primers 5′-CCTCCTATAAGCCTCAGAGCT-3′ and 5′-GGGTGCTTAGTTGGGGGGTGGGATATGAAGGAGGTGCTTGGG-3′. These two PCR products were engineered with overlapping tails that could be annealed together to create a chimeric intron1 with the region deleted. This chimera was amplified with the primers 5′-GGTTTGGGTGGCTAGCCATGGAA-3′ and 5′-CCTCCTATAAGCCTCAGAGCT-3′, digested with I and I, and subcloned into the plasmid (Cushman et al. ) to create the desired transgenic construct. This was subcloned into pBluescript SK+ by an R1 partial digest.
region into the I site of .
The 584-bp region was amplified from the 9.5-kb genomic clone with primers to engineer dIII sites at the ends the sequence (5′-GTCTGGAAGCTTGCTGGTGAGGCTG-3′ and 5′-GGAAGCTTGTCTTGGAGAAGAGACCTCCTCCTGG-3′) and subcloned in both the forward and reverse orientations into the dIII site of the pDeltaODLO 02 plasmid (Iniguez-Lluhi et al. ) obtained from Dr. Jorge Iniguez-Lluhi (University of Michigan, Ann Arbor, MI) that contained the alcohol dehydrogenase () minimal promoter and the firefly luciferase reporter gene. The 508-bp region was PCR amplified from the 9.5-kb plasmid with primers that engineered RI sites at the ends (5′-CGGAAGAATTCTGGTTGCCCAAGGTCC-3′ and 5′-GCCACTCGCAGAATTCATTTC-3′) and subcloned into pBluescript (Stratagene) at the RI site. The was then digested with I/I and subcloned onto a version of pGL3basic (Promega) that contained the minimal promoter inserted into the I/dIII sites. The region was released from this plasmid by digestion with I and subcloned into the I site of pDeltaODLO 02 in both the reverse and forward orientations. The 1196-bp region was amplified from the 9.5-kb plasmid (5′-GGAGTACTGGGACCCTTAAGGCCCTTGGGCTGCAGG-3′ and 5′-GGAGTACTGGAGTCTGAGACAGGAAGACTGAGAG-3′), cloned in to the pGEM-T Easy vector (Promega), digested with I, and subcloned into the I site of pDeltaODLO 02 in the forward and reverse orientations.
Monkey fibroblast CV-1 cells (American Type Culture Collection, Manassas, VA), mouse pituitary gonadotrope αT3-1 cells (Dr. Pamela Mellon, University of San Diego, La Jolla, CA), rat anterior pituitary GH3 cells, and mouse pituitary corticotrope AtT-20 cells were maintained at 37°C/5% CO in Dulbecco’s Modified Eagle Medium (Invitrogen, Carlsbad, CA) supplemented with 10% heat inactivated fetal bovine serum (Hyclone, Logan, UT) and 100 units/ml penicillin-streptomycin (Invitrogen). The GH3 and AtT-20 cell lines were obtained from Dr. Audrey Seasholtz (University of Michigan, Ann Arbor, MI).
Cells were plated onto 24-well plastic plates (Fisher Scientific, Fair Lawn, NJ) at a density of 0.4 × 10 cells/well for CV-1, 0.7 × 10 cells/well for αT3-1, 1.0 × 10 cells/well for GH3, and 1.2 × 10 cells/well for AtT-20 cells, such that cells were 40%–60% confluent the next day. DNA cocktails totaling 0.3 μg/well [0.08 μg enhancer construct, 0.218 μg pBluescript SK+, 0.002 μg (cytomegalovirus) luciferase (Promega) internal control in 400 μl serum-free DMEM] were transfected into cultured cells using Fugene 6 (Roche, Indianapolis, IN) at a 12:5 ratio according to the manufacturer’s protocol. The pDeltaODLO 02 plasmid that contains the minimal promoter with the firefly luciferase reporter gene was used as a negative control and determined as basal level. Forty-eight hours after transfection, Dual-Luciferase Reporter Assay (Promega) was performed according to the manufacturer’s protocol and measured using the Lmax Micro plate Illuminometer (Molecular Devices, Sunnyvale, CA) with the SOFTmax Pro software (Molecular Devices). The results were normalized to the luciferase internal control. All assays were done in triplicate and the results were repeated a total of three times. Results were averaged and expressed as percent activity over basal.
To generate the BAC transgenic mice, BAC RP23-250I22 (supplied by Pieter J. de Jong, Childrens Hospital Oakland Research Institute) was purified over a Nucleobond AX column (BD (Biosciences) and injected into pronuclei of fertilized eggs generated from a cross between DF/B-
males of mixed genetic background (Buckwalter et al. ) and (SJL/J × C57BL/6J) F females. BAC transgenic;
mice were crossed to N6-B6-
mice (Nasonkin et al. ) to generate BAC transgenic;
offspring. These mice and all littermates were weighed, photographed, and genotyped at weaning.
The presence of the BAC was assessed by PCR using primers designed to amplify products that span the junction between the BAC backbone and the mouse genomic DNA (Sp6 end 5′-CATATTTTCCCCATCCACCACCAT-3′ and 5′-TTCCCGCAAGAGCAAACACAAC-3′; T7 end 5′-CCGGAAGGAGCTGACTGGGTTGA-3′ and 5′-TGGGCATTGAGCTTTCTGGGTTTT-3′).
allele (Cushman et al. ) and null allele (Nasonkin et al. ).
The transgenic mouse lines TgN() and TgN() (Cushman et al. ; Vesper et al. ) have been maintained in the mouse facility at the University of Michigan. Newborns were obtained by mating transgenic males of the D4 or D6 transgenic lines with C57BL/6J females (The Jackson Laboratory, Bar Harbor, ME). Genomic DNA was prepared from tail biopsies of all progeny born and then screened for the transgene as previously described (Cushman et al. ).
To create transient transgenic mice with various plasmids, inserts were released from the plasmid vector sequences and purified for microinjection. The 7.7-kb fragment was generated by digestion of the plasmid with I/I. The 8.0-kb fragment was generated by the digestion of the plasmid with I/I. Both inserts were isolated by agarose gel electrophoresis and purified with the Nucleospin Extract Kit (Clontech, Mountain View, CA). Microinjection and transplantation were performed as previously reported (Cushman et al. ). Genomic DNA was prepared from tail biopsies of all progeny born and then screened for the transgene using the same genotyping strategy as for the (Cushman et al., ).
The 7-kb fragment was generated by digestion of the plasmid with I. The 8.5-kb plasmid was generated by digestion of the plasmid with I/I. Microinjection and transplantation were performed as described above. To detect both transgenes, a 250-bp product was amplified from the genomic DNA using the -specific primer (5′-GTGAGAAAACAGGTATCTAGCT-3′) and the -specific primer (5′-CCACTTTGCGTTTCTTGG-3′). Reactions were performed for 33 cycles of PCR conditions: 93°C for 3 min × 1, (94°C for 30 sec, 55°C for 45 sec, 72°C for 20 sec), 72°C for 5 min.
Embryos were harvested on e12.5 from surrogate mothers carrying transient transgenics and quick frozen on dry ice. Cryosections of 12–15 μm were prepared on slides and fixed in 0.5% glutaraldehyde, 1.25 mM EGTA, 2 mM MgCl, and PBS (pH 7.2) for 5 min at room temperature, washed three times in 0.02% NP-40 (Amersham Pharmacia Biotech, Inc., Piscataway, NJ), 100 mM sodium phosphate, 2 mM MgCl wash buffer, and stained for β-galactosidase activity overnight at 37°C in a solution of 1 mg/ml X-gal (Roche), 5 mM KFe(CN), 5 mM KFe(CN), 2 mM MgCl, and 0.02% NP-40 in PBS.
P1 heads from , , , and nontransgenic controls; e12.5 embryos from , , and controls; or e12.5/e14.5 embryos from wild-type animals were harvested and fixed for 2-24 h in 4% paraformaldehyde in PBS (pH 7.2) followed by PBS wash, dehydration in a graded series of ethanol, and paraffin embedding. Six-micrometer sections were prepared on slides and washed in 0.3% Triton X-100 (Sigma, St. Louis, MO) in PBS (pH 7.2) for 15 min at room temperature, permeablized by proteinase K digestion (0.8 μg/ml in 100 mM Tris-HCl, 50 mM EDTA pH 8.0) for 15 min at 37°C, followed by a 5-min fixation in 4% paraformaldehyde in PBS (pH 7.2). To acetylate sections, tissues were exposed to 0.1 M triethanolamine, 0.25% acetic anhydride solution for 10 min. Tissues were prehybridized in hybridization buffer [50% formamide, 5× SSC, 2% blocking powder (Roche Molecular Biochemicals), 0.1% Triton X-100 (Sigma), 0.5% CHAPS (Sigma), 1 mg/ml yeast tRNA, 5 mM EDTA (pH 8.0), and 50 μg/ml heparin]. Tissues were then hybridized overnight at 55°C with the either the or probe diluted in hybridization buffer.
The riboprobe was generated as previously described (Cushman et al. ). The riboprobe was generated by subcloning the polyA region from a modified version of the pnlacF plasmid into pBluescript SK+ (Stratagene) at the HI and II sites (Peschon et al. ). The clone was linearized by digestion with HI to generate the antisense probe. The and riboprobes were generated and labeled with digoxignenin (Roche Molecular Biochemicals) following standard procedures (Mannheim ).
Nontransgenic P1 pituitaries were analyzed for GSU expression with a polyclonal rabbit anti-rat GSU antibody (1:1800; National Institute of Diabetes and Digestive Kidney Diseases, Torrance, CA) and detected with a biotin-conjugated anti-rabbit IgG (1:400; Vector Laboratories, Burlingame, CA) using the Vectastain ABC kit (according to manufacturer’s protocol; Vector Laboratories).
PIT1 immunohistochemistry was preformed on 6-μm paraffin sections of dissected pituitary tissue as described (Charles et al. ).
p
sec
We sought to identify -acting DNA sequences important for mouse expression because regulation of is important for normal pituitary development and function. To accomplish this, we obtained genomic sequence from lemur and pig () BACs and compared these to sequences available online (Ahituv et al. ; Aparicio et al. ; Boffelli et al. , ; Nobrega and Pennacchio ; Williams et al. ). We also generated genomic sequence for the first time for five different primate species to include in the comparison. We identified three conserved noncoding elements (CE) that are larger than 100 bp with greater than 75% identity between human and mouse and tested them for function in cell culture and transgenic mice. The three regions that fit these criteria are , a 300-bp region in the promoter proximal region; , a 200-bp region within the first intron of ; and , a 103-bp region within the 3′ flanking sequence.
Transfection of cultured cells has been successful for demonstrating the function of some elements (Nishimura et al. ; Surinya et al. ; Swamynathan and Piatigorsky ), but there are examples of important regulatory sequences that are not identified with this approach (Lang et al. ; Lettice et al. ; Nobrega et al. ; Zerucha et al. ). is expressed in a distinct spatial-, temporal-, and tissue-specific fashion during development, and the endogenous gene is not expressed in any of the available pituitary cell lines. Nevertheless, exhibited orientation-dependent activity in all cell lines. , located within intron 1 of , also appeared to have orientation-dependent enhancer activity in CV-1 cells and GH3 cells, although at a much lower level. There was no enhancer activity in either the AtT-20 line or the αT3-1 line. The putative regulatory element had low enhancer activity in only the CV-1 and AtT-20 cell lines. There are many possible explanations for the weak activity of and in cell culture, but the cell culture assays did detect enhancer function. The pituitary-derived cell lines were developed in the different hormone-producing cell lineages, e.g., the αT3-1 cells are gonadotrope-like, whereas GH3 cells are somatotrope-like. Therefore, the differential enhancer activities of the different constructs in these cell lines may be examples of context-specific activity. In addition, the cell lines may be more representative of differentiated cell types and, because is expressed significantly only during early pituitary development in the rodent (Sornson et al., ), these cells may not contain the transcription factors and cofactors necessary for the full activity of the putative enhancers. These enhancers also are orientation-specific. However, other examples of orientation-dependent enhancers have been reported (Cheng et al. ; Falvo et al. ; Nishimura et al. ; Surinya et al. ; Swamynathan and Piatigorsky ; Wei and Brennan ).
Our transgenic experiments shed some light on the function of the region. In the context of the promoter, the element results in dorsal expansion of transgene expression. Although this construct used a heterologous promoter, which allows for the expression of the transgene after the endogenous expression is extinguished, these results provide evidence that the region in intron 1 of is important for spatial expression. The region in conjunction with the promoter will be useful for driving the expression of transgenes in the more dorsal aspects of the developing pituitary. Other studies have shown that the Rbp-Jκ DNA binding protein, which is the primary mediator of Notch signaling, can directly bind to intron 1 of and is important for the maintenance of expression (Zhu et al. ). Taken together, these data suggest an role for the in the regulation of expression.
Two kilobase pairs of the promoter proximal region () is inadequate for reporter gene expression in transgenic mice, even in the context of This indicates that additional sequences are necessary for expression in mice.
mice demonstrates that all of the elements necessary for transcriptional regulation of are contained within the BAC. We predict that the remaining critical sequences for expression are within the region immediately surrounding , within 15 kb upstream and approximately 26 kb downstream, because there is a disruption in gene order between human and mouse or pig. These critical control sequences are not readily identifiable by genomic sequence comparisons.
In summary, we identified a region, contained within intron 1 of , which is necessary and sufficient for the spatial expression of in the context of a heterologous pituitary specific promoter. While additional regulatory elements remain to be identified by other approaches, the intronic element is worth screening for mutations in unexplained cases of MPHD patients, especially those that appear heterozygous for mutations in . |
Bones serve many critical functions, including joint movement, ambulation, and vital organ protection. Facilitating these functionalities requires that bone be mechanically stiff, strong, and tough. Although most individuals build bones that are functional for daily activities, a large fraction of these individuals sustain fractures during extreme loading events such as intense physical exercise or falls (Cummings and Melton ; Milgrom et al. ). A major determinant of this fracture risk is bone size. Having slender bones (i.e., small width relative to length) has been associated with increased risk of fracture in children (Chan et al. ; Landin and Nilsson ), young adult athletes and military recruits (Beck et al. ; Crossley et al. ; Giladi et al. ; Milgrom et al. ), and the elderly (Albright et al. ; Duan et al. , ; Gilsanz et al. ; Kiel et al. ). The reason why slender bones are functional for daily activities but perform poorly under extreme load conditions remains unclear. The increased fracture incidence has generally been attributed to the reduced load-carrying capacity of smaller structures (Beck et al. ; Milgrom et al. ). However, recent data indicated that slender bones are also accompanied by matrix-level variations that deleteriously affect tissue quality (Tommasini et al. ). This suggests that there are important interactions between morphologic and tissue-quality traits that may contribute to this clinical problem.
Because most physical bone traits show a high degree of heritability (Leamy ; Susanne et al. ), novel strategies aimed at reducing fracture incidence may be developed by knowing how genetic variation affects the overall mechanical function of bone. Given our understanding of how mechanical function is achieved in bone (Fig. ), at least two major issues need to be incorporated into genetic analyses. First, whole-bone mechanical function is defined by the joint contribution of traits specifying size and shape (i.e., morphology) and traits specifying tissue-level mechanical properties (i.e., tissue quality), the latter traits being defined by matrix composition and organization. Second, anecdotal evidence suggests that there are strong, biological processes that ensure the suite of morphologic and tissue-quality traits generates whole-bone mechanical properties that match daily loading demands (Currey ; Frost ; Olson and Miller ). Traits that covary to satisfy a common function are considered to be functionally related or functionally integrated (Cheverud ; Wright ). Although quantitative trait loci (QTLs) regulating complex properties like bone strength, fragility, and bone mineral density (BMD) have been identified (Beamer et al. ; Klein et al. ; Li et al. ; Orwoll et al. ; Yershov et al. ), rarely have studies been conducted with knowledge of the relationships among genes, cellular processes, growth patterns, physical traits, and mechanical functions (Leamy et al. ; Li et al. ; Li et al. ; Mohan et al. ; Yershov et al. ). Because prior work focused primarily on morphologic integration (Leamy et al. ; Olson and Miller ; Wright ), the effects of variable tissue quality on organ-level function is unclear. Consequently, the identity of the traits that are functionally related and the manner in which these relationships define the repertoire of whole-bone stiffness, strength, and toughness are not fully understood.
Traditional reductionist approaches, because they relate individual bone traits with QTLs, are not useful for this level of analysis because they do not consider how the traits together define mechanical function. Rather, a systems approach is needed to test how variability in whole-bone mechanical properties arises when multiple physical bone traits (or gene sets) vary simultaneously. A viable option is to use path analysis, which is a powerful, multivariate method that analyzes covariances among traits, rather than mean values, in order to reveal functional relations among component traits within complex systems (Wright ). Path analysis has been used to study a variety of complex systems, including bone (Li et al. ; Wright ) and rheumatoid arthritis (Li et al. ). Because path analysis reveals how traits covary in the context of other traits within the system, this approach can be used to identify functional interactions among traits that would not be expected, especially for traits that are defined by diverse sets of genes or biological processes. Importantly, the relationships determined within the multivariate system often differ from the relationships determined from bivariate regression analyses (Grace ). Thus, path analysis, because it provides an accurate representation of functional interactions and deterministic relationships among traits, can be used to understand how gene-level variation leads to increased risk of fragility fractures in bone (Li et al. ).
The goals of this study were to identify bone traits that are functionally related and to determine how these functional interactions contribute to variability in whole-bone stiffness, strength, and toughness. We studied these interactions using inbred mouse strains because different strains show widely varying skeletal traits (Jepsen et al. ) and because transverse growth patterns of mouse long bone, which defines bone slenderness, are similar to human long bone (Garn ; Price et al. ). Importantly, because different inbred mouse strains build mechanically functional bones by assembling different sets of physical bone traits during growth (Jepsen et al. , ; Tommasini et al. ; Turner et al. ; Wergedal et al. ), inbred mouse strains provide a valuable model to study how interactions among traits define mechanical functionality. We examined two particular inbred mouse strains, A/J and C57BL/6J (B6), because a biomechanical analysis revealed that A/J mice have more slender femoral diaphyses but thicker cortices and higher mineralization when compared with B6 mice, which have wider femoral diaphyses but thinner cortices and lower mineralization (Jepsen et al. ). Surprisingly, femora from the two strains showed similar stiffness values, suggesting that there are interactions among bone size, cortical thickness, and mineral density and that these interactions are important for building a functional bone. However, A/J femora failed in a more brittle manner compared with B6, indicating that these strains provide a valuable model to investigate why genetic variations that affect bone slenderness also affect bone fragility.
To test the hypothesis that bone size, cortical thickness, and mineral density are functionally related, we conducted a path analysis using data derived from a panel of AXB/BXA recombinant inbred (RI) mouse strains. RI strains have a unique pattern of genetic randomization that can be used to measure the tendency for different traits to cosegregate (i.e., correlate) in a natural, nonpathologic manner rather than to map genes (Nadeau et al. ). For long bones like the femur, each RI strain will show a unique suite of adult traits (Fig. ), depending on how the particular set of genes for each strain influenced the cellular processes regulating bone growth (Price et al. ). Certain bone traits are postulated to covary so that organ-level functionality (i.e., adequate whole-bone stiffness) is achieved for each RI strain. Because the size, shape, and tissue quality of the femoral diaphyses will differ among the RI strains, a correlation analysis conducted across the RI panel should thus reveal which traits covary. If the interactions among bone size, cortical thickness, and mineral density are part of a basic biological paradigm that facilitates the development of organ-level functionality, then we would expect to see these particular traits covary across the RI panel. To test how the functional interactions among these bone traits define whole-bone mechanical properties, we conducted additional path analyses that included whole-bone stiffness and toughness and developed causal models based on engineering principles.
AXB/BXA RI mice derived from A/J and C57BL/6J (B6) progenitor strains were examined in this study. Male and female A/J, B6, and 20 AXB/BXA RI strains ( = 9–17/genotype/sex) were bred at The Jackson Laboratory (Bar Harbor, ME, USA) and shipped to the Mount Sinai School of Medicine (New York, NY, USA) at 3.5 weeks of age. Including males and females allowed us to test whether dimorphic bone growth patterns lead to sex-specific interactions among traits. The handling and treatment of mice was approved by the Institutional Animal Care and Use Committee. To standardize environmental conditions, mice were fed a standard rodent chow (Purina Rodent Chow 5001) and water , subjected to a 12-h light:dark cycle, and raised with approximately 5 mice/cage in the same room. Mice were killed at 16 weeks of age because previous studies showed that growth-related changes in traits slowed prior to this age (Price et al. ). Femora were harvested and stored frozen in phosphate buffered saline at –20°C. Femoral length (Le) was measured from the proximal femoral head to the distal condyles using digital calipers (0.01-mm resolution).
Diaphyseal cross-sectional morphology and tissue mineral density (TMDn) of the femur were measured using an eXplore Locus SP Pre-Clinical Specimen MicroComputed Tomography system (GE Healthcare, London, Ontario, Canada). Three-dimensional images of the entire femur were obtained at an 8.7-μm voxel size. The analysis region was limited to a 2.5-mm region of the mid-diaphysis that was located immediately distal to the third trochanter (Fig. ). This site corresponded to the location where most femora failed during the four-point bending tests (see below). Femora were individually thresholded using a standard thresholding algorithm (Otsu ) to segment bone and nonbone voxels. A custom analysis program (The Mathworks, Inc., Natick, MA, USA) was developed to quantify morphologic traits describing the amount of tissue (cortical area, CtAr; marrow area, MaAr; total area, TtAr; cortical thickness, CtTh) and the spatial distribution of tissue (moment of inertia, J). Moment of inertia is a measure of the proximity of the tissue to the geometric centroid of the cross section. The amount and distribution of tissue are both necessary to properly relate diaphyseal morphology to mechanical function, because bones having the same cross-sectional area but different moments of inertia (e.g., a solid cylinder and a tube) will exhibit different mechanical behaviors under bending and torsional loads (van der Meulen et al. ). Total bone area was defined as the sum of the cortical and marrow areas. The relative cortical area (RCA = CtAr/TtAr) provided a measure of the proportion of the total area that was occupied by bone. These traits were quantified for each cross section and the values were averaged over the volume of interest. The morphologic parameters measured by micro-computed tomography were found to be within 1% of histologically determined values for an independent set of adult AJ, B6, and C3H/HeJ femora (data not shown).
The microCT images were also used to quantify tissue mineral density (TMDn). TMDn is the average mineral value of the bone voxels only and was expressed in hydroxyapatite (HA) density equivalents. TMDn was calculated by converting the gray-scale output of bone voxels in Hounsfield units (HU) to mineral values (mg/cc of HA) through the use of a calibration phantom containing air, water, and HA (SB3: Gamex RMI, Middleton, WI, USA). TMDn was defined as the average bone voxel HU value divided by the average HU value of the HA phantom multiplied by 1130 mg/cc (HA physical density). The same calibration phantom was included in all scans to adjust mineral density measurements for the variability in X-ray attenuation inherent to independent scan sessions. Validation studies using 44 mouse femora showed that tissue mineral content correlated linearly ( < 0.01) with both ash weight/hydrated weight and ash weight/dry weight (data not shown).
Following microCT analysis, femora were loaded to failure in four-point bending at 0.05 mm/sec using a servohydraulic materials test system (Instron Corp.; Canton, MA, USA) to assess whole-bone mechanical properties (Jepsen et al. ). Load deflection curves were analyzed for stiffness (the slope of the initial portion of the curve), maximum load (Max Load), postyield deflection (PYD), and work-to-failure (Work). PYD, which is a measure of ductility, was defined as the deflection at failure minus the deflection at yield. Yield was defined as a 10% reduction of secant stiffness (load range normalized for deflection range) relative to the initial (tangent) stiffness. Work, which is a measure of toughness, was defined as the area under the load deflection curve. Femora were tested at room temperature and kept moist with phosphate buffered saline during all tests.
Bone traits for male and female AXB/BXA RI strains were compared using a two-way analysis of variance (ANOVA) to test whether mechanical properties and physical bone traits were inherited in a sex-specific manner, similar to prior studies (Orwoll et al. ). A direct comparison between sexes was facilitated by converting trait values to scores to minimize size effects.
where is the trait value for each mouse femur and and SD are the mean and standard deviation, respectively, calculated using the average values for all 20 AXB/BXA RI strains. This transformation standardizes the variables so each trait shows a mean of zero and a standard deviation of one. Female and male values were -transformed separately so that the phenotype of a female mouse was compared with that of other females and the phenotype of a male mouse was compared to that of other males.
To test whether bone traits covary, Pearson correlation coefficients were calculated for all trait-trait comparisons. This analysis used the mean scores for each RI strain in the panel. The correlation matrix retained the magnitude and direction (positive, negative) of each correlation coefficient. Statistically significant correlations were identified by establishing a threshold correlation magnitude. The correlation threshold was determined using permutation tests (Churchill and Doerge ; Nadeau et al. ), which corrects for multiple comparisons and establishes the maximum correlation coefficient that arises when the bone traits are randomly arranged across the RI panel.
A path analysis was conducted using the mean scores of each RI strain to determine how functional interactions among bone traits contribute to variability in whole-bone mechanical properties. Causal models were constructed by specifying the directed path between select bone traits. Directed paths identify related traits and indicate the direction of the causal relationship between them. Three causal models were constructed in order of increasing complexity. The first causal model (Fig. ) was constructed to test the hypothesis that variability in bone size (TtAr) was causally related to variability in CtTh and TMDn. Because the causal path between CtTh and TMDn is not known , we tested paths going in both directions. Femur length (Le), which is correlated with body weight, was included to take body size into consideration and to determine how variability in cross-sectional size (TtAr) relative to length (i.e., slenderness) relates to CtTh and TMDn. Males and females were tested separately, rather than using sex as a categorical variable, to generate two independent sets of path coefficients.
The second causal model added two variables to test how the functional interactions defined in the first model contributed to whole-bone stiffness (Fig. ). Stiffness, which is a measure of the amount of deflection a bone undergoes while loaded, was used in this model because most theories suggest that bone adapts to daily loading demands by adjusting physical traits to keep peak tissue-level strains (deformations) within a certain range (Frost ). An advantage of using the femoral diaphysis in this analysis is that the mechanical behavior of cylindrical structures is well characterized. Cortical area (CtAr) was selected as the second variable because engineering principles state that stiffness depends on a measure of bone size and a measure of tissue quality, which was already represented in the first model (i.e., TMDn). The causal paths in the second model flow from the physical bone traits toward stiffness, since mechanical properties are the outward manifestation of the underlying traits.
The third model (Fig. ) added two complex mechanical properties (PYD and work) to the prior models. These two mechanical properties capture the failure process of bone and thus differentiate whether a bone fails in a brittle (low PYD and work) or ductile (large PYD and work) manner during extreme loading events such as an overload condition.
Path coefficients, which represent the magnitude of the direct and indirect relationships between traits, were calculated based on the hypothesized causal models and the variance/covariance matrices of the observed data. Structural equations were constructed using the path coefficients to specify the interconnected, causal relationships. Analyses were run for males and females separately using the standardized (-transformed) data (LISREL v. 8.8; Scientific Software International, Lincoln Park, IL, USA). Observed and model-implied covariance matrices were compared using maximum likelihood estimation and overall fit was determined by a chi-squared test. Unlike conventional null hypothesis testing, path analysis favors the , theory-based model such that models are rejected only if the observed data and the expectations derived from the model do not match (i.e., if < 0.05) (Grace ). Thus, chi-squared (χ) values with an associated value greater than 0.05 means that the data are adequately fit by the model. The root mean square error of approximation (RMSEA), which takes the number of degrees of freedom of the model into consideration (MacCallum and Hong ; Stieger and Lind ), was also reported as an additional fit index. For RMSEA, < 0.05 indicates close fit, 0.05 < < 0.08 indicates fair fit, and > 0.10 indicates poor fit (MacCallum and Hong ).
The mean trait values and the standard deviations for each AXB/BXA RI strain are shown in Table (females) and Table (males). The mean values were normally distributed across the RI panel for all bone traits ( > 0.1, Kolmogorov-Smirnov test). Representative femoral cross sections of female RI strains (Fig. ) illustrate how the size of the femoral diaphyses ranged from being smaller than A/J to larger than B6. The male RI panel (not shown) showed the same variation as the females. The overall mean trait values, which were determined by averaging across the RI panel (bottom row in Tables and ), were intermediate between the A/J and B6 parental strains for all traits.
Trait values for female and male RI strains were compared to test whether bone traits were inherited in a sex-specific manner.
values for the linear regressions ranged from 0.65 to 0.87 ( < 0.0001 for all regressions). Figure shows representative regressions for a morphologic trait (total bone area) and a complex mechanical property (postyield deflection). Male RI strains tended to be heavier than their female counterparts, as expected, and this translated into male mice showing larger morphologic traits compared with those of females. Despite the differences in body size, a comparison of the scores between male and female RI strains showed significant effects due to genotype ( < 0.0001, 2-way ANOVA), but not to sex ( > 0.9, 2-way ANOVA). Thus, the data indicated that male and female AXB/BXA RI strains inherited bone traits in a similar manner.
A correlation matrix was established to identify the traits that cosegregated (i.e., correlated) in a significant manner (Table A and B). For 20 AXB/BXA RI strains and 12 traits, the permutation test indicated that a correlation coefficient of 0.66 corresponded to a significance level of < 0.1, a correlation coefficient of 0.69 corresponded to < 0.05, and a correlation coefficient of 0.75 corresponded to < 0.01. Plotting the mean traits across the AXB/BXA RI panel revealed that many traits cosegregated in a significant manner for males and females (Fig. ). Of the 66 correlations analyzed, females showed 27 strong correlations ( > 0.66, < 0.1) and 22 of these were considered significant ( > 0.69, < 0.05) (Table A). Males showed 17 strong correlations ( > 0.66, < 0.1) with 13 considered significant ( > 0.69, < 0.05) (Table B). The average number of strong correlations per trait was 4.5 for females and 2.6 for males, indicating that bone traits were highly connected or interdependent. Networks depicting the significant trait interactions are shown in Fig. .
The first path model (Fig. ) had no available degrees of freedom to properly assess goodness of fit.
values. The structural equations for TMDn and CtTh were significant ( < 0.01) and 45%-56% of the variability in TMDn and CtTh was explained by Le and TtAr for the male and female data sets. The path coefficients linking CtTh to Le and TtAr were similar for females and males, suggesting that variability in body size and bone cross-sectional size had similar effects on cortical thickness for both sexes. Importantly, the path coefficient between TtAr and CtTh was negative, indicating that a decrease in bone size (i.e., a more slender bone) was associated with a thicker cortex. The path coefficients for TMDn were also similar for females and males and indicated that CtTh and TtAr were significant predictors of TMDn. The structural equations indicated that when holding bone length fixed, a mouse showing a 1-SD decrease in TtAr (i.e., more slender bone) would also show an increase in CtTh by approximately 0.2 SD for females and males. Because TMDn was influenced by both TtAr and CtTh, the 1-SD reduction in TtAr would be associated with a 0.42-SD increase in TMDn for females [−0.27 (direct path) + −0.23 × 0.67 (indirect path) = −0.42] and 0.34-SD increase in TMDn for males [−0.21 (direct path) + −0.16 × 0.79 (indirect path) = −0.34]. Thus, the net effect of a unit change in TtAr on TMDn was similar for both sexes. These results indicated that mean trait values covaried among the RI strains in such a way that larger bones (i.e., larger outer diameter) tended to have thinner cortices and lower mineral content, whereas smaller bones (i.e., smaller outer diameter) tended to have thicker cortices and higher mineral content. Thus, the analysis of the RI panel indicated that there are important functional interactions among bone size (TtAr), cortical thickness (CtTh), and mineral density (TMDn).
The second model (Fig. ), which added CtAr and stiffness (Stiff) to the first model, showed a good fit for both males and females as determined by the χ and RMSEA goodness-of-fit indices. Path coefficients were similar for both sexes, and 98% of the variation in cortical area (CtAr) was explained by TtAr and CtTh. The weak path coefficient for bone length indicated that length influenced CtAr indirectly through TtAr and CtTh. Stiffness was positively related to CtAr and TMDn for both females and males. The combination of a morphologic trait and a tissue-quality trait explained 70%-85% of the variation in stiffness for males and females.
In the third model (Fig. ), adding PYD and work to the prior two models did not affect the goodness of fit as determined by χ and RMSEA for either males or females. PYD was positively related to CtAr but negatively related to TMDn. These two traits explained 40% of the variation in PYD for females but only 20% for males. Work was positively related to both stiffness and PYD, and 88%–89% of the variation in work was explained by these two mechanical properties.
The path analysis confirmed that the cross-sectional size of cortical bone, cortical thickness, and tissue mineral density were functionally related and determined that these functional interactions contributed to variability in whole-bone mechanical properties. Prior studies also reported correlations among bone traits (Ferretti et al. ; Jepsen et al. ; Tommasini et al. ; Turner et al. ; Wergedal et al. ), but interpretations of these interactions were based largely on an intuitive sense of how bone works. Path analysis provides a rigorous, statistical method that explains how bone traits interact and how these interactions define complex mechanical properties. The covariation among traits observed in the RI panel indicated that the functional interactions among morphologic and tissue-quality traits are part of a basic biological paradigm that allows for flexibility in how organ-level functionality is achieved in mouse long bone. Thus, gene variants that affect bone slenderness may be accommodated by the covariation of tissue quality and vice versa. The functional interactions observed in the path models also explain why certain traits and mechanical properties show pleitropic effects (Yershov et al. ). The current data do not provide insight into whether variation in bone slenderness is causal to variation in mineral density or the opposite is true. The fact that the direction of the directed path between CtTh and TMDn had no effect on the model suggested that these traits covary with each other in a functional manner (i.e., are co-adapted).
A principle finding of this study was that significant correlations were observed among many bone traits (22 for females and 13 for males) and that each trait was related to, on average, three to five other traits. These significant correlations support the hypothesis that many bone traits are functionally related and share common biological controls affecting growth (Wright ). In contrast to single-gene perturbations, which often create pathologic conditions (Bonadio et al. ) and in some cases elicit strong, adaptive responses (Bonadio et al. ), this analysis used data derived from AXB/BXA RI strains to study how perturbing multiple genes simultaneously, in a nonpathologic manner, affected the construction of adult long bone (Nadeau et al. ). The results supported the premise that genetically randomizing genomic regions would result in each RI strain building a mechanically functional femur, but in slightly different ways, depending on the particular set of genes that were inherited from A/J and B6. None of the RI strains showed fractures, suggesting that each RI strain achieved organ-level functionality, i.e., a healthy bone. The randomization of A/J and B6 genomic regions was associated with a large range of trait values among the 20 AXB/BXA RI strains, and for several RI strains trait values exceeded (were larger or smaller) those of A/J and B6. This was expected given that these bone traits are genetically complex.
The correlations among stiffness, maximum load, PYD, and work-to-failure and those among stiffness and maximum load and the morphologic traits like CtAr and J were expected because they are consistent with engineering principles (van der Meulen et al. ). The only major discrepancy between the current study and our prior work (Jepsen et al. ) was that the negative correlation between postyield deflection (PYD) and tissue mineral density was not statistically significant. Because tissue ductility has been shown to depend on mineral, collagen, and water content (Currey ; Martin and Ishida ; Wang et al. ), the weak correlation between PYD and TMDn highlights the need to expand the repertoire of matrix compositional traits to find more meaningful tissue-quality relationships. This is particularly important because small variations in TMDn are correlated with large changes in bone stiffness, strength, and ductility (Currey ).
The path analysis provided an understanding of how variability in whole-bone mechanical properties arises from genetic variation in the underlying bone traits. The causal models, which were developed based on engineering principles and empirical data describing bone growth patterns (Price et al. ), showed good fits for both the male and the female data sets. The various trait substitutions and path additions/deletions had little effect on the model, suggesting that the relationships among traits fit the engineering-based causal models in a robust manner. The functional interactions among traits specifying cross-sectional bone size, cortical thickness, and mineralization indicated that more slender bones (smaller TtAr relative to length) were compensated by thicker cortices and higher tissue mineral density, whereas wider, more robust bones (larger TtAr relative to length) were compensated by thinner cortices and lower tissue mineral density. The fact that perturbing the interactions among these traits resulted in inadequate fits for the models suggested that the functional interaction between morphology and tissue quality was a fundamental biological process that allowed each RI strain to achieve an appropriate whole-bone stiffness during growth. Although the data do not reveal the details of the biological processes or the genes responsible for these functional interactions, the fact that the physical traits of each RI strain covaried in such a way that adult bones were sufficiently stiff and strong for loading demands suggested that these biological processes were adaptive in nature (Frost ).
The path analysis revealed novel interactions between traits that do not have an obvious relationship (e.g., TMDn and morphology). The positive correlation between cortical thickness and TMDn suggested that there is coordinate biological regulation between the amount of mineral packed into the matrix (TMDn) and the relative movements of the periosteal (TtAr) and endosteal (MaAr) surfaces that define all morphologic traits, including cortical area, cortical thickness, and moment of inertia. A similar relationship between cortical thickness and mineral density was reported previously (Wergedal et al. ). Total area (TtAr) was treated as an exogenous (independent) variable in the causal models because this particular trait reflects the movement of the periosteal surface. Cortical thickness provides a measure of the relative expansion of the endosteal surface, which defines the size of the marrow cavity (i.e., MaAr). These correlations argue strongly for a high degree of biological control over the final multivariate product. This coordinate regulation of traits is consistent with the concept of morphologic integration described five decades ago (Olson and Miller ). Although prior studies focused largely on the genetic and biological mechanisms regulating bone morphology (Leamy et al. ; Richman et al. ), the current data indicated that variation in bone morphology is also linked to variation in matrix composition and thus to tissue quality (Swartz et al. ). This analysis, which was conducted using a single bone from a single species, is consistent with prior comparative analyses that examined bones serving different functions from different species (Currey ). Thus, the interaction between bone morphology and tissue quality appears to be an important biological paradigm for bone. Further studies need to be conducted to determine if the functional interactions identified for AXB/BXA RI femora hold for different bones or for different intercrosses and outbred populations.
Although this study was conducted using genetically randomized mouse strains, there is no reason to expect that functional interactions are limited to multigenetic variation. If the functional interactions among traits are a basic biological paradigm, then environmental perturbations or single-gene mutations that alter cross-sectional bone size would be expected to also perturb tissue quality and vice versa. For example, a mutation affecting type I collagen synthesis was associated with reduced tissue strength and a compensatory age-related increase in bone size (Bonadio et al. ). Although the data do not reveal the limits to the amount of variation that can be accommodated by the underlying biological processes, the skeletal dysfunction associated with certain genetic mutations (e.g., osteogenesis imperfecta) and environmental perturbations (e.g., scurvy) clearly tell us that not all genetic or environmental variations can be accommodated by the functional interactions among bone traits. Understanding these limits and how to perturb the environment to facilitate a positive response may provide new targets for genetic analyses as well as new strategies for building more robust bones during growth.
Although the trait sets for each RI strain appeared to achieve the appropriate stiffness and strength for day-to-day activities, not all sets of traits resulted in satisfactory values for PYD or work-to-failure (e.g., AXB5, AXB13, BXA26). These latter mechanical properties are important because they reflect the failure process of bone. Brittle failures are associated with low values for PYD and work-to-failure. A brittle femur would be expected to perform poorly under extreme loading conditions such as the cyclic loading associated with intense physical exercise or the high-impact loads associated with falls. Although PYD and work-to-failure are defined by several matrix compositional and organizational traits (Wang et al. ), the amount of mineral packed into the matrix (i.e., the mineral:matrix ratio) is particularly crucial because this trait is positively related to tissue stiffness but negatively related to tissue ductility (Currey ). Having higher TMDn may help compensate for smaller cross-sectional bone size by increasing the stiffness of the cortical tissue, but this comes at the expense of reduced ductility (decreased PYD) and reduced toughness (decreased work-to-failure). Consequently, the functional interactions between morphologic traits and tissue-quality traits creates preferred sets of traits for bone such that a wide bone (large TtAr) with low TMDn is preferred over a slender bone (small TtAr) with high TMDn.
Preferred sets of traits are not limited to the mouse skeleton; in fact, similar relationships between bone size and tissue-level mechanical properties have recently been reported for the human skeleton (Tommasini et al. , ). Young adult males and females with slender tibiae were found to have a compensatory increase in tissue-level stiffness and a concurrent reduction in tissue-level ductility and damageability when compared with individuals with wider tibiae. The variation in tissue stiffness and ductility appeared to arise primarily from variation in ash content, similar to what we found for the mouse skeleton (unpublished data). This variation in tissue ductility may contribute to the increased incidence of stress fractures observed for young adult athletes (Crossley et al. ) and military recruits (Milgrom et al. ) having narrow bones. These studies suggest that functional interactions among morphologic and tissue-quality traits are similar for the mouse and human skeletons.
The similarity in scores between female and male RI strains indicated that males and females inherited bone traits in nearly identical manners. Our results differed from prior work that reported that whole-body BMD was inherited in a sex-specific manner for BXD RI strains (Orwoll et al. ). Further studies need to be conducted to resolve whether this discrepancy in sex-specific heritability is a function of the particular intercrosses (AXB/BXA versus BXD) or the nature of the traits that were examined (specific measures of morphology and tissue quality versus whole-body BMD). The path coefficients describing the functional interactions among bone size, cortical thickness, and tissue mineral density were similar for females and males, suggesting that sexual dimorphism did not appreciably affect the relationship among these traits. Furthermore, males and females showed similar path coefficients for the more complex models (Models B and C) indicating that both sexes showed similar relationships between mechanical properties and the underlying physical bone traits. This may help explain why none of the traits were inherited in a sex-specific manner.
#text |
Type 2 diabetes mellitus (T2DM), along with associated problems such as hypertension, dyslipidemias, and obesity, is an increasing problem in human populations: 150 million individuals are currently affected and a rapid expansion is expected during the next 20 years (Freeman and Cox ). Human genetic studies provide strong evidence that predisposition to T2DM, and in particular its defining phenotype, glucose intolerance, has a complex genetic basis (McCarthy ). This is supported by studies in animal models. A significant number of candidate genes for involvement in predisposition to T2DM are implicated in pancreatic β-cell dysfunction (Freeman and Cox ). Glucose-stimulated insulin secretion (GSIS) is the pivotal homeostatic process in the control of blood glucose levels and takes place in pancreatic β cells (Ashcroft and Rorsman ). GSIS takes in a number of relatively well characterized biochemical pathways such as glycolysis, the TCA cycle, and the respiratory chain (Fig. A), but new discoveries implicate novel features of β-cell biochemistry (Eto et al. ; Newgard et al. ; Ronnebaum et al. ; Rubi et al. ). The effects of individual mutations on the GSIS system are not always intuitively obvious and qualitative descriptions do not allow for a quantitative analysis of these effects. It should be possible to overcome these problems by building explicit mathematical models of the system.
Systems biology aims at system-level understanding of biological processes and how a system’s behavior emerges from the interactions among its components. An objective milestone for successful cell simulation might be the construction of a whole metabolic model. Consequently, biochemical dynamic models composed of a relatively large number of metabolic reactions are being developed. Examples are models of central carbon metabolism in (Chassagnole et al. ; Varner ), glycolysis in lactic acid bacteria (Hoefnagel et al. ), mitochondrial NADH shuttles and anaplerosis in β cells (Westermark et al. ), and mitochondrial ATP production (Bertram et al. ; Magnus and Keizer , , . The advantage of such detailed, biochemically formulated models is that a one-to-one comparison can be made between model and experiment. Thus, they provide platforms that allow discovery of new intrinsic biological properties.
A number of mathematical and computational models have been developed related to the GSIS system. Topp et al. () designed a model that includes three ordinary differential equations (ODEs) representing the dynamics of glucose and insulin within the mass of β cells. As a coarse-grid model, the model concentrates on investigating the normal behavior of the glucose regulatory system and pathways into diabetes. Another minimal model was developed by Bertram et al. (). They built a simplified ATP synthesis model based on earlier models of oxidative phosphorylation (Magnus and Keizer , , to capture the same behavior as in Magnus and Keizer’s models. In the Bertram et al. model, they took pyruvate as the main input of their system and ATP as the end product. Four mitochondrial variables (NADH, ADP, ΔΨ, and Ca) are described with equations corresponding to the dynamics of different types of fluxes or reactions. With their simplified model they investigated the dynamics of the four mitochondrial variables versus the change of glycolytic flux and pulses of calcium. A more refined simulation model for the mitochondrial system was developed by Yugi and Tomita (). In this model 58 enzymatic reactions and 117 metabolites are included to represent four pathways (respiratory chain, the TCA cycle, fatty acid β oxidation, and the inner-membrane transport system) in mitochondria. Previously published enzyme kinetics studies from the literature were integrated into a single dynamic model using the E-cell2 simulation platform ().
Following recent research on the importance of NADH shuttles in ATP production, a mathematical model of mitochondrial NADH shuttles has been developed by Westermark et al. (). The model comprises the mitochondrial NADH shuttles and mitochondrial metabolism, which in total included 19 enzyme reactions and 10 metabolites with simplifications on the boundary of the system. The model reproduces the experimental finding by Eto et al. () that blocking the NADH shuttles attenuates the signaling to ATP production while retaining the rate of glucose oxidation. Besides the model of NADH shuttles, Westermark and Lansner () also constructed a model of the upper part of glycolysis in the pancreatic β cell. The model concentrated on studies on the enzymatic reactions of PFK (phosphofructokinase), which has been shown to control metabolite oscillations in the glycolysis pathway. Although more and more computational dynamic models have been constructed, a complete, detailed kinetic model with a large number of enzyme rate equations mimicking the dynamics from glucose input to ATP production has not been developed so far.
We therefore initiated the development of a quantitative, kinetic model of the core processes in GSIS as an aid to understanding genetic and biochemical data in mouse models of T2DM. Constructing such a model poses a significant technical and intellectual challenge because the GSIS system consists of at least five biochemical subpathways (glycolysis, TCA cycle, the respiratory chain, NADH shuttles, and the pyruvate cycle) that take place in more than one cellular compartment (cytoplasm and mitochondrion). In addition, the system consists of a large number of interacting metabolites and enzymes. Here we describe a kinetic model of the pathways leading from the beginning of glycolysis to ATP production, which we call the core process. We show that the model has qualitative properties consistent with expectations, including showing oscillations in the glycolysis pathway and in ATP concentration. We also discuss some analyses of the properties of the network and its applicability to mouse models and the understanding of human disease.
We used mathematical modeling in which all metabolite interactions are described in terms of kinetic equations. The components of our mathematical model are kinetic parameters and state variables, which indicate the state of a system at a certain time. The kinetic parameters include Michaelis-Menten constants, rate constants of association and dissociation, etc. Most of the kinetic parameters were taken from the literature (see Supplementary Materials).
Our model is based on ODEs and consists of 44 enzymatic reactions, 59 metabolic state variables, and 272 parameters. The number of parameters is large because most of the rate equations inside the mitochondrion follow complex reaction mechanisms (Ping-pong Bi Bi, Ordered Bi Bi, etc.), although identical reactions in different compartments have been assumed to have the same parameters. The following is a representative derivation of one of the 59 ODEs. To determine the change in the concentration of a certain metabolite [
] over time, we calculate the sum of the reaction rates producing [
] minus the rates consuming [
]. Thus, we get the following differential equation, where represents one of the 59 metabolites:
The simulation is then equivalent to solving the following differential equations:
in which represents the parameters in the model.
The most basic dynamic property of the GSIS system is oscillation along glycolysis (see Westermark and Lansner ) and the concentration relationship between ATP and glucose (Detimary et al. ). Figure shows dynamic simulation results of some metabolites in the network. The metabolites along the glycolysis pathway show the desired oscillations. The metabolites outside the glycolysis pathway (e.g., malate and citrate inside the mitochondrion) did not show oscillations. ADP and ATP showed oscillations, as required for stimulation of Ca transport in β cells (Westermark and Lansner ). ATP concentration increased with the change of glucose concentration from 0 to 20 mM, while ADP concentration decreased. Figure shows the concentration change of ATP with changes of glucose concentration. Consistent with the observations of Detimary et al. (), the model predicts ATP levels to increase more than twofold between 0 and 10 mM glucose and then plateau. At 6 mM glucose, the ATP level was 1.78-fold higher than the initial state concentration. ADP progressively decreased to 0 mM as glucose increased from 0 to 6 mM (data not shown). It should be noted that the ATP concentrations predicted by the model represent an upper bound on intracellular ATP concentration because they do not take into account consumption of ATP by K channel activity and other cellular processes.
MacDonald () reported that murine pancreatic islets do not contain detectable levels of malic enzyme (malate dehydrogenase; ME), whereas human and rat islets do. We therefore simulated the effect of removing ME from the model. The model showed a 7.5% increase in cytoplasmic malate concentration and a 9.6% decrease in pyruvate on deletion of ME, as well as small effects on the concentrations of some other metabolites (of the order of 1%–2%). However the resultant ATP concentration was not significantly altered. The oscillatory behavior of the model was also unaffected.
#text
xref
fig
#text
With a detailed biochemically formulated model, it is convenient to analyze the behavior of any metabolite of interest within the system. However, for analysis of the global properties of the system, the model could be simplified. Sensitivity analysis-based methods can be used to identify those parts of the model that are relatively unimportant for the properties of the system, and those parts could be eliminated from the model (Dano et al. ). Sensitivity analysis also provides an indication of the biological realisticness of the model. We therefore performed sensitivity analysis of the parameters along each subpathway using SBaddon and the Systems Biology Toolbox (Schmidt and Jirstrand ). Figure A shows the analysis results for the glycolysis pathway. The rest of the results are shown in the Supplementary Materials. We can see that along the glycolysis pathway parameters in reactions v1, v2, v6, and v8 have the most effect on the state variables. Three reactions in the TCA cycle also showed a high level of sensitivity: v9, v14, and v17. The other subpathways in the model showed less sensitivity to parameter changes than glycolysis. Typically, any sensitivity observed lay at the beginning of a particular subpathway. Thus, sensitivity was observed at reaction v9 at the beginning of the TCA cycle (from the perspective of incoming pyruvate), at reaction v17 in the NADH: malate aspartate shuttle, which is the reaction consuming incoming malate, at reaction v24 in the NADH:glycerophosphate shuttle, which is the first step of the respiratory chain, and at reactions v41 and v44 of the pyruvate cycle.
Correlation analysis is based on the output from sensitivity analysis and represents concerted behavior of parameters in the system. Correlation analysis of glycolysis (Fig. B) shows strong interreaction correlations between most of the reactions in the pathway with the exception of reaction v8 (pyruvate transport across the mitochondrial membrane), which showed only self-correlation. Strong cross-correlation was also observed in the pyruvate cycle. In the other subpathways, particularly the TCA cycle and NADH:malate aspartate shuttle, little evidence of correlation was seen between reactions. However, correlation was seen only in the NADH:glycerophosphate shuttle between reactions v28, v40, and v43 (import of ADP into the mitochondrion, production of ATP by complex V, and its export) and reactions v38 (the start of the NADH:glycerophosphate shuttle, catalyzed by G3PD [glycerol-3-phosphate dehydrogenase]) and v4 (catalyzed by GAPD [glyceraldehyde-3-phosphate dehydrogenase]).
In this article we describe a mathematical model of the core dynamics of the glucose-stimulated insulin secretion (GSIS) pathway of pancreatic β cells. This model draws on a body of previous work on the modeling of the glycolysis pathway (Nielsen et al. ) and mitochondrial metabolism (Yugi and Tomita and references therein) but brings in additional features, notably NADH shuttles and the pyruvate cycle (Eto et al. ; Newgard et al. ; Ronnebaum et al. ; Rubi et al. ), to provide a more comprehensive view than has previously been available. The model calculates the changes over time in the concentration of 59 metabolites. We have assessed the behavior of the model with respect to two core properties of the GSIS system: the relationship between ATP output and glucose input concentrations and oscillations of metabolite concentration in the glycolysis pathway. By measuring the nucleotide content of islets isolated from NMR1 mice, Detimary et al. () showed a twofold increase in ATP content between 0 and 10 mM starting glucose concentration, with a subsequent flattening off of the rate of increase. Our model shows the same pattern of ATP concentration change (Fig. ). Oscillations in the glycolysis pathway are a well-established feature of the system (Westermark and Lansner ) and are observed in our model for a number of metabolites in this pathway but not for metabolites outside glycolysis (Fig. ). Metabolites showing oscillatory behavior in our simulations do include ATP and ADP, which have been shown to show coordinated oscillations with glucose-6-phosphate concentration, calcium concentration, and insulin secretion (Deeney et al. ; Nilsson et al. ).
Figure summarizes investigations designed to map the parameter ranges within which oscillations in the glycolysis pathway are observed in our model. This study identifies the reactions catalyzed by PFK, PGP, and PK (reactions v2, v5, and v6 in Fig. B) as being particularly sensitive to parameter changes, while reaction v3 was particularly insensitive to parameter changes. PFK was previously identified as being central to the oscillatory behavior of this pathway and in particular for modulating the frequency of oscillations (Westermark and Lansner ), but roles for PGP and PK are not well established and merit further investigation.
We also investigated the global properties of our system. Choke point analysis (Rahman and Schomburg ; Yeh et al. ) identifies reactions that either uniquely consume a specific substrate or uniquely produce a specific product. These are suggested to be particularly sensitive points in metabolic networks (Yeh et al. ). We identified six enzymes—GK, PFK, FBA, PGP, PK, and NDK—as choke points. Five of these are in the glycolysis pathway, the exception being NDK, suggesting that defects in the glycolytic pathways are likely to be particularly deleterious to insulin secretion. Intriguingly, the first of these enzymes, GK, has been found to be mutated in Maturity-Onset Diabetes of the Young, Type II (MODY 2) (Vionnet et al. ). This raises the possibility that choke point analysis of particular systems may help identify candidate genes.
Sensitivity analysis of the glycolysis pathway (Fig. A) indicates that four reactions (v1, v2, v6, and v8) are sensitive to changes in parameter values. Three of the five choke points in the glycolysis pathway are also identified by sensitivity analysis as important. Reaction v1 is catalyzed by GK, which has been argued to be the primary control point of glycolysis, i.e., it controls the flux through the pathway (Matschinsky et al. ; Sweet and Matschinsky ; Wang and Iynedjian ). V2 is the PFK reaction discussed above, while v6 is the PK reaction, which is also a choke point on the glycolytic pathway. V8 is the transport of pyruvate into the mitochondrion, whose effects are notably not correlated with the rest of the glycolytic pathway. Outside the glycolytic pathway the model is relatively resistant to perturbation. However, it does show significant sensitivity in reactions v9 of the TCA cycle (the first step after pyruvate import), v17 of the NADH:malate aspartate shuttle (the first reaction consuming imported malate), v24 of the NADH:glycerophosphate shuttle (first reaction of the respiratory chain), and v41 and v44 of the pyruvate cycle. Therefore, these may be important control points in the network.
Taken together, these properties of the GSIS core model are consistent with previous experimental and modeling results. It should be emphasized, however, that the concentrations of metabolites that our model predicts have not been tested against experimental evidence. In addition, the parameters of our model, although based on published values and producing acceptable qualitative behavior, have not been optimized as a concerted set. That is, although we have considered the effects of altering individual parameters on the behavior of the system, we have not explored parameter space in a consistent way to estimate an optimal set of values or range of values (e.g., Liebermeister and Klipp ). To do this we will require a set of experimental measurements tailored to the optimization of the model. The value of generating such a data set will lie in testing our ability to test the relationships between measured metabolite concentrations and in the model optimization per se.
The utility of this model is expected to lie in its ability to aid our understanding of processes leading to type II diabetes (although it may also have other uses). To do this we will need to extend it to include additional processes, most notably the generation and detoxification of reactive oxygen species, which has recently been implicated in a mouse model of type II diabetes (Freeman et al. , , nd downstream processes leading to the transport of calcium in response to fluctuations in ATP concentration and the eventual export of insulin.
p |
It will not be long before we have mutant alleles for every gene in the mouse genome. It also will not be long before we can sequence an entire genome in a few hours and for less than 1000 US$. But there is much more to come, and the speed with which these developments will take place will surprise you. Imagine following the fate of every single cell during the development of a mouse from fertilization to birth and even beyond. Imagine watching the expression of a single molecule of any protein or the total expression of all proteins in a single cell continuously over time. Imagine titrating the expression of single genes in specific cell populations at will.
High-throughput technologies have become the driving force in the analysis of biological systems. Biologists are increasingly taking advantage of automatization, miniaturization, and computerization. In this sense biology follows the development of computer and information technology: smaller size, higher speed and capacity, lower cost. However, we should remember that the age of computer and information technology was preceded by a pre-exponential phase during which important theoretical frameworks and concepts were developed: Alan Turing () and John von Neumann () provided the mathematical basis for an automatic computing machine and a corresponding “computer architecture.” To Claude E. Shannon (Weaver and Shannon ) and Norbert Wiener () we owe the mathematical theory of information and cybernetics. The convergence of electronic and mechanical engineering then triggered the development and application of systems control theory, a key requirement for the modeling and simulation of the dynamics of technical systems.
A major challenge in biology is to model, simulate, and eventually predict the behavior of complex biological systems. The identification of the individual components that constitute a biological system, i.e., through the genome-wide transcriptome, proteome, and metabolome analysis, will be required but will not be sufficient to achieve this goal. We will also need detailed information about the “network architecture” and the dynamics of biological systems. This is where systems biology comes into place (Kitano ; Kirschner ; Palsson ; Alon ).
Biological systems are emerging, adaptive systems, highly complex and often nonlinear. Their behavior cannot be explained solely on the basis of their individual parts. Deep insight into the network structure, function, and dynamics of biological systems can be obtained only through their systematic perturbation, followed by a detailed characterization of the molecular, cellular, and phenotypic changes that follow these perturbations. Based on the perturbation consequences observed, a model can then be established or existing models modified or further developed that grasp the important features of the underlying mechanisms (Sauro and Kholodenko ).
Mouse genetics has been an extremely powerful perturbation method for nearly a century. Loss-of-function and gain-of-function mouse mutants are able to reveal causal relationships between specific genes and specific phenotypes. An impressive mouse genetics toolbox is now available that allows us to perturb a wide range of biological systems. Methods such as the production of transgenic mice, gene targeting through homologous recombination in embryonic stem (ES) cells, phenotype or gene-driven mutagenesis strategies, or RNAi-based knockdown are now used on a routine basis. Sequence diversity is also a form of natural perturbation. In combination with the analysis of gene expression and phenotype analysis, a thorough comparison of the consequences of allelic variants can be very powerful. The main challenge will be the functional dissection of the combinatorial activity of small sequence changes, forming the core of “Complex Trait Analysis.”
Perturbing biological systems through genetic changes is only one way to obtain information to dissect the structure and function of genetic networks. Equally important and increasingly appreciated is the use of small molecules, which can act as agonists or antagonists of biological processes (Schreiber ). Whereas a while ago combinatorial chemistry was largely a domain of pharmaceutical drug development, the power of small molecules as a means to study the function of specific proteins or pathways is increasingly appreciated. There are specific strengths and weaknesses to the use of small molecules. One of the most important aspects is specificity. Very rarely does a small molecule bind to one and only one target; in many cases the precise number and nature of targets is unknown. Small molecules are more adaptable to the titration of dose-response or pulse-chase studies. Similar to searching for modifiers in a genetic screen, chemical biologists now are starting to perform combinatorial screens to unravel redundancies or pathway interactions that are not revealed by single small-molecule screens. In fact, one may be able to stay below the “toxic window” of a specific small molecule by combining two or more of them, each of which acts on different targets within the same pathway. Eventually we will see a convergence between the fields of small molecules and small animals, i.e., in the area of noninvasive imaging (Sako ). Molecular markers will become available that allow us to follow perturbations at the molecular and cellular level and in real time.
#text
A common theme in advanced technologies and engineering is to divide systems into modules that can be treated individually or in terms of connecting different modules as part of a higher-order system. Since the discovery of the double-helix structure of DNA, a reductionist approach to analyze biological systems has proved to be extremely successful. However, we feel that we are reaching a limit as to how much we can learn about complex biological systems by looking with increasing resolution at individual components of a system. No doubt, at the end we would like to understand biological phenomena on the basis of atomic resolution. On the other hand, the rise of systems biology reflects our increased appreciation and desire of looking at all the scales of biology, including the molecular, cellular, organismic, and population-based levels.
Partitioning biological systems into modules helps to achieve a more integrated picture. To understand causal relations among individual parts and modules, we need information about the directionality of flow of information or material between the edges within a network (Natarajan et al. 2006). We already know that systems behave differently depending on whether we deal with one or a few molecules or millions of molecules. Stochastic and statistical approaches, i.e., Bayesian network reconstruction algorithms, need to be applied to deal with the uncertainties and probabilities of biological systems (Needham et al. ). The role of noise in biological systems is just being unraveled. Some of the most important contributions are currently made by physicists who are able to apply the repertoire of statistical physics to biological problems (Rao et al. ; Samoilov et al. ; Sprinzak and Elowitz ; Kussell et al. ; Alon ).
Given their complexity, a remarkable feature of biological systems is their robustness with respect to environmental perturbations (Kitano ,; Kitano and Oda ; Kurata et al. ). How do biological systems preserve their function despite environmental conditions that can differ over magnitudes of scale leading to tremendous fluctuations in metabolic components or ligands? We do not yet understand the underlying mechanisms that are responsible for this robustness. Genetic redundancy, i.e., the presence of multigene families that can at least partially substitute for each other, is apparently one way to increase the robustness of a system. Similarly, a redundancy of pathways could contribute to the potential of a cell to maintain the robustness of a biological system. On the other hand, there might be a price to be paid, i.e., under different environmental conditions, leading to a tradeoff of robustness versus fragility dependent on the external factors that act on the system (Kurata et al. ). Robustness or fragility of biological systems can be understood only if we obtain insight into the structure and the dynamics of elements responsible for feedback control, an essential element in almost all complex systems (Schmidt and Jacobsen ).
italic
#text
Maybe we have focused too much on the similarities of model organisms instead of also trying to understand the differences. Maybe we should increase our efforts in comparative systems biology. We might have to take a much closer look at the differences between mice and humans in terms of their relevance for drug development and try to understand the mechanisms of species-specific absorption, distribution, metabolism and excretion (ADME). Some of the species differences can be overcome by, for example, introducing human genes into the mouse genome or by xenografting human stem cells into mice (Shultz et al. ). These efforts in “humanizing mice” are still at the “trial and error” stage. We urgently need a comparative systems analysis that could guide us in selecting the most relevant genes or cell types that are the cause of differences in drug responses or disease pathogenesis and that should be prioritized in our efforts to improve the predictability of mice as a model system for human disease.
Biological systems are complex adaptive systems that emerge during the development from a fertilized egg to the development of an adult organism. During evolution changes in the environment lead to different constraints and fixation of certain degrees of freedom in genome structure and function. Components of genome networks can be added or changed only when the workability and functionality of the biological system is maintained, at least to a certain degree (Ottino ; Weitz et al. ).
Comparative systems analysis needs appropriate databases (Albeck et al. ; Kersey and Apweiler ). These are not yet sufficiently developed. The mouse comparative ontology database () is useful but does not provide information about the components, interactions, and dynamics of physiologic systems.
We need all the information available, i.e., a user–friendly, easily retrievable information system on the level of transcripts of a given cell, the dynamic response of mouse vs. human cells to small molecules, the levels of redundancy in the two species, species-specific genes, splicing patterns, and post-translational modification.
#text
Systems biology often tries to apply formal mathematical descriptions based on time-series analysis of biological response. So far the sheer amount and the quality of data constituted significant roadblocks to tackle the dynamics of biological systems. Technological advances help us to overcome these problems. A more severe problem, at least for the current generation of biologists, is the limited training in mathematics. The first two years of engineering training provides the mathematical toolbox necessary for a mathematical description of technical systems and is essential for modeling or simulating the behavior of complex systems. It will be neither possible nor useful to turn every biologist into a mathematician. However, we need to improve the dialog between biologists and mathematicians, physicists, and engineers. The basics of linear algebra, vector analysis, and graph theory have to enter the curriculum of a biologist’s training (Wingreen and Botstein ).
Unfortunately, formal tools for model production do not yet exist. In addition, model building is not easy and requires a very good understanding of the biological system under study. A question often raised is where to start: bottom up, top down, or a combination of both. An interesting suggestion is to start “middle out,” where the modeling begins at the level at which there are rich biological data and then reach up and down to other levels (Noble ). Another major difficulty is the transfer of a model from one application to another. We need to develop standardization frameworks so that even novices in computational biology or systems biology are able to build, access, and work with existing models (Wall et al. ).
For more than 100 years mouse genetics has relied on the analysis of single monogenic mutants. The methods to identify or produce mutants have changed considerably over the years. Soon we will have in our catalogs and freezers mouse mutants for every gene in the genome (Collins et al. ). Extensive collections will also be available as a result of phenotype-driven mutagenesis screens (Balling ). Whereas the analysis of these mutants might keep us busy for many years to come, the next frontier of mouse genetics is already on the horizon: systems genetics. We all know that the expressivity and penetrance of mouse mutant phenotypes can vary tremendously, depending on the genetic background. Modifier screens can be used to identify some of the genetic loci responsible for the strong influence of genetic background on physiologic and pathophysiologic processes. Sequencing and, as a cheaper substitute, SNP typing have provided us with a detailed picture of the genetic diversity of our main inbred mouse strains. Most of them are derived from a very limited pool of parental strains, and strong selection was applied to obtain the handsome, highly adapted common lab strains of mice that we now use in our experiments.
Recombinant inbred strains and other reference panels of inbred strains are powerful tools for performing a genome-wide dissection of complex biological traits that are the result of multiple, quantitative, and often highly interacting genes (Churchill et al. ; Flint et al. ; Zou et al. ; Hill et al. ; Peters et al. ). The series of BXD strains has been a paradigm for the success of analyzing complex traits. Unfortunately, the use of recombinant inbred strains does not fall under the category “quick and easy” but requires a fair amount of logistics, infrastructure, and an appreciation for the power of genetics. The major bottleneck, however, was the “power of mapping resolution” that the analysis of 30-80 recombinant inbred strains provides. The Complex Trait Consortium has tackled precisely this problem (Churchill et al. ). The goal is to produce approximately 1000 recombinant inbred strains (The Collaborative Cross) within the next five years and make them available as an open source to the scientific community. Importantly, the parental strains chosen include three strains that we would classify as “inbred wild mice,” i.e., PWK/PhJ, WSB/EiJ, and CAST/EiJ. The inclusion of these genetically highly diverse strains adds about 75% additional sequence diversity. The availability of this large panel of diverse and well-structured strains will allow experiments where mice with an identical genotype can be produced in large numbers and compared to an equally large number of mice with a wide range of different genetic and even environmental backgrounds. Sequencing of the parental strains and a community-based complementary and additive phenotyping will eventually produce a resource that will help us to answer questions about gene function, epistatic genetic interactions, and genome-environment interactions that we can currently only dream about.
There are other approaches, i.e., the development of consomic mouse strains, that essentially target the same questions (Peters et al. ). It will be important to not look at these approaches as exclusive or competitive, but as a new toolbox of quantitative trait analysis where each one has specific pros and cons. New phenotyping methods, including gene expression arrays, or phenotyping based on noninvasive imaging will have to be integrated into the described complex trait studies. Microarrays are a new microphenotyping platform that allow us to look at the expression of thousands and hundreds of thousands of different genes (eQTLs). This shift to microphenotypes requires new statistical tools because of multiple-testing issues but it also gives a much higher computational capacity then ever before. To quote Denis Noble: “Biology is set to become highly quantitative in the 21st century. It will become a computer-intensive discipline” (Noble ).
For many years mouse genetics has been the driving force as a hypothesis generator for functional genomics. Mouse models, i.e., transgenic mice, knockout mice, or mouse mutants identified from phenotype-driven screens, are great tools to identify candidates for human disease genes. The construction of mouse inbred strain panels derived from genetically diverse parental populations provides us with valuable model populations. At the same time, the power of human association studies has reached a point where some people even think that it heralds the end of mouse genetics. I think the opposite is true. The availability of mouse reference populations will allow us to ask questions that complement those addressed by human association studies. More importantly, we can quickly validate hypotheses derived from human population studies not only by constructing equivalent mouse populations but also by probing the function of individual genes through the analysis of gene targeting or specific point mutation alleles. The argument that we can find such mutations also in human populations does not take into account that in mice we are not only able to study the effect of genetic variation, but also to “titrate the environment” much better than it will ever be possible for humans.
At this time, mouse geneticists and human geneticists have not connected well enough to exploit the power of their respective toolboxes. To quote Rob Williamson: “There is still an impedance mismatch between human association and reductionist mouse studies.” Maybe this special issue of can contribute to better cooperation between mouse and human geneticists. It will pay off for all of us. |
Up until the 1980s, most research in developmental biology involved analyzing the interactions among and within the tissues that participated in some embryologic event (e.g., limb development) and, on the basis of careful experimentation, inferring something about these interactions. A second and complementary approach was to use kinetics and other theoretical approaches to model a problem in development such as patterning. In either case, where there was more than one possible explanation of a phenomenon, it seemed obvious and sensible to give preference to the explanation that seemed the most parsimonious on grounds of natural selection. The gradual and continuing discovery of the intricacy of the signalling conversations between participating tissues, the richness of the activated molecular networks that regulate developmental change, and the complexity of the resulting processes have shown just how naïve was that original paradigm.
Over the past two decades, our ability to use a wide range of molecular technologies to investigate these regulatory networks and to collate the patterns of gene expression characterizing a particular state of differentiation has produced enormous amounts of information, often accessible from online databases (e.g., ), on how development proceeds. This ability to exploit the new technologies and so to explore complex developmental events at the molecular level has enabled the field, over a period of some 20 years, to progress from a small-scale subject interesting relatively few scientists to an area of major interest and excitement across the world. One stimulus here has been the realization that mutation-derived errors in these networks underpin many human congenital abnormalities. The consequent study of these abnormalities, often using mouse models, has the dual benefit of advancing medical research and giving us a tool to pry open these networks. A second has been the realization that homologous networks do similar things in very different organisms and that we therefore have a means to explore the mechanisms of evolutional change which usually operate, as Waddington was probably the first to emphasize, through mediating changes in development (see below and Waddington ).
All this work has led to a wonderful increase in our understanding of developmental events, particularly those that involve signalling and those in which the activation of a transcription factor initiates a new process (for review, see Gilbert ). That said, it has to be admitted that, for most developmental events, there are now large amounts of molecular expression data that are hard to interpret unambiguously. Often we do not really know in a particular event which proteins are important, which are secondary, and which are background, and knockout and other experimental data can be either ambiguous or unhelpful. In one sense, the situation is worse than it was in the 1980s: Then we could appeal to parsimony via natural selection to make choices; now things are so complicated that we have no means of recognizing parsimony, and would not trust the concept anyway.
One approach to this complexity is to say that if only we had enough data, everything would become clear, but it is unlikely that anyone in the field really believes this. A second is to say that we need better and stronger intellectual frameworks than just relational databases for organizing and analyzing the new data that are pouring out of laboratories. A third is to take the view that we need not just a better framework for handling data, but better intellectual ideas. The third view is certainly right, but those ideas have yet to emerge and, in the absence of some deeply original and intuitive thinking, may well emerge from the second approach, which is hard enough at the moment and which is articulated under the general name of . This approach is new and does not yet have any formal structure but, in general, seeks to embed the events of a particular developmental event within a computational and hierarchical framework that links tissues, cells, processes,and molecular/genomic data, and often aims to capture the results of high-throughput technology (e.g., Kimelman ). Perhaps the best-known example of a systems approach is the work on sea-urchin development (e.g., Ben-Tabou de Leon and Davidson ), which integrates tissues, genes, and networks (Longabough et al. ). Other important systems approaches include analyses of developmental networks (Xia et al. 2006) and the molecular basis of very early mouse development (Eviskov et al. ).
This article does not seek to provide a systems approach to any particular phenomenon but to consider how best to take advantage of the computational tools currently available so as to ensure that systems descriptions based on tissues, cells, and processes can be interoperable in the sense that they use a common language. This would enable them to query one another and use each other’s formal knowledge (much as we can do for genes and proteins that are already linked through their IDs to their appropriate database). The key tools here are ontologies and the purpose of this article is to discuss what ontologies are, how they can be used in formalizing systems approaches to development within and across species, and what are the resulting benefits.
At the core of development is the predictable production of functional and differentiated tissues from early, less well-defined tissues. It would therefore be sensible if, when one person uses, for example, the term “E14.5 mouse left atrium” in his systems model of heart development, another person using the same term in her model can link to that of the first. The way that such linkage is done for proteins is to use an ID from a standard database (e.g., the protein ID from Uniprot, ), and because proteins are all amino-acid strings and hence of the same rank, they can readily be stored in the tables of relational databases.
Anatomical tissue organization, in contrast, is hierarchical in nature: The vertebrate hindlimb, for instance, is obviously partitioned into regions (thigh, knee, calf, foot), each of which has its own parts, and the concept of “hindlimb” would naturally be expected to include these subordinate parts, together with information about their relationship to the hindlimb and to one another. While it is obviously straightforward to assign a unique ID to a given tissue at a given developmental age, it is clear that the hierarchical organization of tissues poses some organizational problems beyond those needed for handling sequence data.
The way that such hierarchical information is most appropriately handled is through These are domains of knowledge formalized in a way that allows them to be computationally accessible. In practice, ontologies are built up by linking in a hierarchical way. Here, a is a triad of the general form <termterm> and terms can have parents and children (e.g., the E14.5 left atrium the E14.5 heart; the E14.5 heart the E14.5 cardiovascular system, etc.). Although they are tedious to produce (even the simplest organ system has a great many tissues and a lot of organization), there are now ontologies for the tissues of all the main model adult organisms and for the developmental anatomy of the mouse, zebrafish, and (accessible from the Open Bio-Ontologies site, ). Every term in these ontologies carries a standard ID of the form <abcd><ijkl>, where abcd gives a short letter code for the ontology (e.g., EMAP for mouse development) and ijkl gives the number for a specific tissue at a specific developmental age (e.g., EMAP:7917 is the ID for the E14.5 mouse left atrium, with EMAP standing for the Edinburgh Mouse Atlas Project, ). It is these IDs that allow for interoperability because they represent defined concepts (or terms) that can be used anywhere, even as synonyms.
It is worth noting that such an anatomical ontology is more than just the list of the parts as it includes a great deal of knowledge about how these parts are organized into larger structures and these larger structures into organ systems (e.g., Fig. ). Such an ontology may also include additional knowledge built on other relationships such as (an ontology of developmental anatomy would well include lineage relationships) and type data (e.g., the femur bone). There is also no reason why a child should have only a single parent in the ontology: For example, it is equally appropriate to describe the femur as <><the skeleton> as <><the hindlimb>, and a rich ontology could well include both relationships (and this multiparenting of terms means that it would be called by the technical term , or ). This is not the place to include a detailed discussion of how anatomical ontologies are built and used (the interested reader should consult Bard ), but it should be mentioned that the internal organization of an anatomy ontology is usually rather complex (the structure needs to be able to handle many relationships as well as definitions and links) and is best read in a browser program such as OBO-Edit or COBrA (Figs. and ; Aitken et al. ; Harris et al. ) that is visualized in a GUI rather than as a list on paper. There are several languages in current use for handling ontologies (the best known are OBO and OWL) and they can be translated into each other using the COBrA tool.
In the context of systems developmental biology and in addition to the appropriate anatomy ontology, there are two general ontologies that are also useful. The first is the Cell-Type Ontology (Bard et al. ) and the second is the Gene Ontology (Ashburner et al. ; Harris et al. ). The former, unlike the anatomy ontologies, not only includes all the common and many of the uncommon cell types that are found across the phyla but it is essentially species-independent and so facilitates cross-species analyses and comparisons. This ontology is structured to include our knowledge of the many properties of these cell types and each is separately coded under function, morphology, ploidy, development, etc., using two relationships, and (see Fig. ). This ontology is thus a terse summary of a great deal of knowledge about cell types and their properties.
The Gene Ontology or GO is by far the best known and most used of the standard bio-ontologies (it is used for protein annotation in Uniprot). Unlike Uniprot, it does not include sequence information but focuses on the properties of proteins and includes hierarchical knowledge about cellular locations, molecular functions, and the functional processes in which they are involved. For systems developmental biology, it is the latter that is the most important and the process hierarchy includes a wide variety of developmental processes (although they are distributed across the ontology rather than integrated under a single heading [Fig. ]), each of which, of course, has a unique ID. Of particular interest here is the database of proteins that is linked to the GO so that a user can easily identify all the stored proteins associated with a GO term, or the GO terms associated with a chosen protein (although it should be said that keeping this database up-to-date is a major task).
One important factor about ontology terms is that they can be associated with data (usually held in a standard relational database and linked to the ontology via the appropriate IDs); examples include the proteins that satisfy the definition of a GO term (), the genes expressed in a particular mouse tissue at a particular time, (), and the micrographs associated with a pathologic state (). Here, the hierarchical knowledge within the ontology comes into play: If, for example, a user requires the genes associated with the developing mouse forelimb at E12.5, the response comes from searching the ontology to identify the constituent tissues in the limb and using their IDs to collect all the associated data. This can be done because this type of relationship has the property known as . This means that if a term has data associated with it, then these data can be associated with the parent (e.g., a gene expressed in the tarsus is also expressed in the hindlimb). Propagation is associated with some bio-ontology relationships (e.g., ) but not with others (e.g., ; one would not expect pigment cells to have the same properties as their neural-crest-cells precursors).
Ontologies, together with their linked data, provide an important online resource and have several key roles in systems developmental biology. The first is the use of well-defined terms (with their associated IDs) to standardize annotation, the second is for linkage to databases that store data associated with the terms, and the third is to facilitate the identification of similar terms in very different contexts. These ideas are explored here and in the next section and, while the approach is applicable to the development of any organism and also across organisms (see Discussion), the examples focus on the mouse. This is because our knowledge about its development is now so deep that it is often possible not only to describe how any tissue develops morphologically over time (Kaufman and Bard ) but to identify the processes and changes in cell type that underpin each time slice of a tissue’s development (see ). In addition, the mouse community is fortunate to have access to substantial online informatics resources that are available from The Jackson Laboratory. In the context of this article, the most important of these is GXD, a database of gene-expression (G-E) data for the developing mouse in which expression data are annotated with (and hence searchable by) tissue name, developmental stage, and GO IDs, as well as other genetic identities.
Although development is complicated, it can be seen as the operation on tissues of relatively few core processes that involvewith each process having subprocesses. Such processes are archived in the Gene Ontology (see above and Fig. ).
If the formation of a system is to be modeled, then the first step is to lay out its normal pattern of development graphically. Much of the stage-by-stage lineage data for mouse embryogenesis is available in text format (Kaufman and Bard ) and can be linked to the tissues (with their IDs) in the ontology of mouse developmental anatomy (Bard et al. ) and hence with GXD. Staging of mouse embryos is based on the appearance of standard external identifying features as embryogenesis proceeds; Theiler staging for the mouse gives, in essence, two stages a day when things are going rapidly (E6–E12.5) and one stage a day when the appearance of new features is slower (E1–E5 and E12.5 onward). Annotating the tissue names is straightforward because each tissue at each stage has a unique ID accessible from the ontology of mouse developmental anatomy (e.g., Fig. ).
For annotating changes in the state of cellular differentiation, there are two options. The first is to use an appropriate GO term, but because the GO includes less than 50 cell types, it cannot (and was not intended) to do justice to developmental anatomy. A better option, therefore, is to use the higher-level GO term “cell differentiation” and combine it with two terms from the Cell-Type Ontology.
In the case of metanephric mesenchyme being induced to form nephron epithelium, the annotation is
or, using the appropriate IDs from the two ontologies:
Where the state of a tissue changes between two Theiler periods, one can annotate the developmental change that drives this transition (this is not the usual way in which development is considered!) with the appropriate GO process terms. In this way, the final graph has, superimposed on the lineage flow of developmental anatomy, the appropriate differentiation and process terms that drive the development of each tissue. Underpinning each of these transitions is the appropriate ontology ID, so that the final graph is set up to be complete, formal, and interoperable.
Figure illustrates the result of annotating in this way the development of the mouse urogenital system over Theiler stages 13–19 (E8.5–E11.5), where classical descriptive embryology has, first, shown that the intermediate mesoderm differentiates to give the nephric duct, mesonephros (from which develops the gonad) and the metanephros, and, second, given the cell types associated with each tissue. The graph also includes the results of experimental work that has clarified the processes that push development from one stage to the next. For clarity, the figure excludes the IDs but they are all readily available from the appropriate ontologies.
There is one immediate use of this model that derives from annotating terms with standard ontology IDs. The graph as it stands has no molecular data, but all the current gene-expression information associated with a particular developing mouse tissue at a given Theiler stage is computationally accessible from GXD through ID interoperability. GXD genes also carry GO IDs which enables searches to be quite sophisticated. It is straightforward, in principle at least, to use these GO IDs to identify, for example, signals and receptors for tissues that signal to one another (Bard ) or transcription factors that are synthesized at a particular stage and ready for a future event.
In short, ontology annotations of developmental systems are not only the key to interoperability and standardization of systems models, they give rich searching possibilities.
There is a further bonus from such annotations: As development proceeds, the same developmental processes are used in very different contexts within one organism. This similarity goes beyond the differentiation of the same cell type from different tissues (e.g., neurons can differentiate directly from neuroepithelium and indirectly after the migration of cells originally from the neural crest or from epithelial placodes). Obvious examples are the branching of epithelial tubules (in glands and in the vascular system), epithelial folding in its many forms, the forming of mesenchymal condensations (the first step in the development of muscles, bones, and cartilage), and the initiation of movement (in tissues as different as neural crest cell, primary germ cells, neurons, and gastrulating epiblast cells), pigmentation (retinal epithelium, neural crest cells). Indeed, such processes are common to development across the phyla.
Consider the hypothesis that each of these processes can be viewed as a “motor” driven by the activation of particular set of transcription factors (TFs). If this hypothesis is correct, then each set of tissues that are about to participate in a particular event should express those TFs and they should be present in the appropriate G-E profile in the associated database (they may also be missing, but they should not have been shown to be absent). If so, then the overlap of the G-E profiles of tissues about to initiate a particular process should include those proteins involved in initiating that process and housekeeping proteins common to all (or at least most) cells. If these housekeeping genes can be excluded, such Boolean analysis should yield key proteins involved in that process. A similar analysis of the G-E profiles for those tissues immediately after they have initiated a particular process should yield those proteins involved in that process.
Any analysis along the lines suggested makes several assumptions beyond that of common TFs underpinning common processes. First, the time resolution of the G-E database has to be fine enough to discriminate between the period of a tissue’s to undergo a process and the process itself. In the case of mouse development for which the database archives expression by Theiler staging, this means time slices of 24 hours for early and late mouse development and 12 hours over the period E6–E12; this is probably adequate. Second, the database needs to contain enough data on the expression of all relevant genes. This latter criterion is unlikely to be met, even for GXD. While this rich resource currently includes some 250,000 expression results for about 7500 genes (information courtesy of Dr. Martin Ringwald), the data are not uniformly distributed across tissues or time slots. There is thus an element of chance as to whether the database holds information about the expression of a gene in a particular tissue at a given time.
A preliminary analysis of the G-E data associated with some process widely undergone during mouse development should thus be based on the following lines:The analysis as a whole should provide a set of candidate genes for the process of interest (provided that the group of tissues as a whole is undergoing no more than one common development-specific process). However, the process is quite lengthy and, given that GXD may include several hundred expressed genes for even a single tissue at a specific Theiler stage, can really only be properly handled through a substantial computational infrastructure that has yet to be put in place. Indeed, the more expressed genes that can be associated with the set of tissues, the more reliable will be the analysis. The situation will get better but more complicated once GXD includes microarray and other high-throughput data.
Fortunately, there is a relatively simple shortcut that can be used for a quick exploration of the approach, and which takes advantage of the GO annotations in GXD. If one merely restricts one’s searches to the periods of competence of tissues about to initiate a process and genes with a GO transcription factor ID, the output should be restricted to those TFs associated with the initiation of that process. As an example, consider the mesenchyme-to-epithelium transition that takes place many times during development. A preliminary examination (full details will be published elsewhere) of the gene-expression profiles in GXD shows that there are substantial entries for the formation of blood vessels in the early mouse heart, the differentiation of heart endocardium, the metanephric ducts, the mesenchyme that forms the mesonephric ducts, and the early stages of somite development. If the search is restricted, using GO IDs, to TFs in the participating tissues in the two Theiler stages before these transitions take place, the data show that three TFs, Lhx1, Foxc1, and Meox1, are present in all these tissues (apart from a couple where there is incomplete data). A further inspection of the complete distributions of these genes shows that their expression (insofar as it is fully represented in GXD) is highly restricted over space and time, and because they do not in general overlap one another, they cannot be considered as housekeeping genes; their coexpression is hence unlikely to be a coincidence.
The TFs Lhx1, Foxc1, and Meox1 are thus, as a set, good candidates for collectively initiating a mesenchyme-to-epithelium transition, although they seem not to have been previously identified as fulfilling this role. It is therefore a prediction that this set be expressed in other tissues undergoing a mesenchyme-to epithelial transition. Examples that might be worth investigating here include the stromal fibroblasts in the cornea that become the corneal endothelium and the splanchnopleure mesoderm that forms mesothelium (GXD currently includes no relevant expression data for these tissues). If the prediction were confirmed, it would be worth investigating which proteins were synthesized following their activation.
Although systems developmental biology is thought of as a relatively new subject, its basis lies in an idea that Waddington originally had in the late 1930s and that he expressed graphically as (Fig. ). In its original form (left), the picture was of a ball rolling down the valleys of a complex hillside. In its final form (right), this picture showed that the topology of this surface was shaped through complex linkage to a set of pegs on an underlying flat surface. The meaning of this metaphor is that the developmental trajectory of a developing cell over time (the rolling of the ball down the valleys) is shaped by its environment (the undulating surface), with the form of the landscape being determined by the interacting properties of many genes (the pegs and their ties). Local development was thus viewed as a gene-based interaction between a particular group of cells and its environment and the system had to be viewed as a whole. This was a startlingly original view to hold more than 50 years ago, at a time when the scientific community almost uniformly held the view that all genes did was code for enzymes and such trivialities as eye color (Van Speybroeck ). Its value as a metaphor was shown by Waddington’s use of it to describe evolution: Small changes (mutations) in genes led to changes in the landscape and hence to altered patterns of development and so to novel organisms (see Waddington ).
We are now beginning to catch up with Waddington’s thinking. Systems models are starting to be produced that aim to integrate the complexity of the molecular, cellular, and tissue details that underpin development using the computational resources that are now available. There will be many more such models, and, given the richness and complexity of development, they are bound to overlap. It is important that such overlaps allow interoperability, and a key point made in this article is that the community should not only use, as it already does, the terms and IDs for the gene and protein databases, but also incorporate the terms and IDs for cells, tissues, and processes that are to be found in standard bio-ontologies. This is partly for interoperability across models, but also to allow direct linkage to such databases as those handling G-E data.
This article also points out that there is an additional bonus from using these IDs, i.e., where the same process is used in the development of different tissues, the linking of tissue IDs to their associated gene-expression profiles can, in principle, lead through Boolean analysis to the identification of candidate genes associated with the initiation and execution of this process. The databases are currently populated with genes whose roles are still unclear so this computational approach complements experimental approaches because it enables small groups of genes to be linked to the initiation and execution of processes. This contrasts with the analysis of individual genes whose roles can be analyzed using, for example, transgenic technology and high-throughput technology that picks up a large numbers of genes but yields little about their function.
A further point to be made is that the type of computational approach to the identification of gene function given here allows us to test the hypothesis that common processes are underpinned by common TFs. If such a search yielded several candidate TFs that are found to be associated with the initiation of a process in some tissues but have yet to be found in all tissues, it suggests that the expression of these TFs should be further examined in these other tissues. A lack of expression there would cast doubt on or at least narrow the extent of the hypothesis. This approach to systems biology thus provides assays for testing our ideas.
In this article, the focus has been on formalizing mouse development and analyzing the molecular underpinnings of its underlying processes because it is for this organism that the associated expression database has the finest spatial and temporal granularity. It should of course be pointed out that the approach is equally applicable to other organisms and even across organisms that share equivalent developmental processes. In the first instance, the mouse can be used as a model for identifying process-associated genes. Where a similar process occurs in other organisms, the homologs will be candidate genes for that process (and the XSPAN facility, , will be helpful in identifying equivalent tissues in model organisms). A further tool under development that may be useful in this context is CARO, the Common Anatomy Reference Ontology (, Haendel et al. 2007) which aims to provide interoperability across species-specific anatomy ontologies. In the longer term, one can envision the construction of complex systems models that span organisms and that employ the full richness of the computational resources that are available.
At a slightly deeper level, what distinguishes systems developmental biology from other approaches to unpicking the complexities of development is the formalization of the events of embryogenesis. This in turn enable tissues, cells, and expressed genes to be linked in a way that lends to computational as well as other forms of analysis. The exercise of formalizing embryogenesis encourages the biologist to think in ways that complement other more traditional approaches. |
Since the derivation of the original inbred mouse strains from populations of fancy mice to investigate the genetic basis of cancer (reviewed in Paigen ), many additional inbred strains have been derived that harbor a tremendous amount of natural genetic variation (Beck et al. ; Ideraabdullah et al. ). However, unlike the more recently produced wild-derived strains, the vast majority of commonly used inbred strains trace their ancestry to the original mouse-fancier populations. An analysis of the genomes of extant inbred strains was recently made possible using data from a 15-strain resequencing project (), which revealed that the most widely used laboratory inbred strains are not random composites of the three main mouse subspecies (, , and ), but have a remarkably high level of shared ancestry largely contributed by the subspecies (Yang et al. ). Since many of the original inbred strains are also the most widely used in biomedical and laboratory research, the architecture of the genetic variation in derived resources is highly dependent on the interconnected and complex breeding histories of the progenitor inbred strains (Lyon et al. ).
Over the last fifty years, numerous genetic resources have been devised and developed for specific purposes using a variety of inbred strains as progenitors (reviewed in Silver ). The major genetic resources that are widely used currently include recombinant inbred (RI) lines (Bailey ; Broman ), recombinant congenic strains (RCS) (Demant and Hart ), genome-tagged or congenic (CON) lines (Iakoubova et al. ), chromosome substitution strains (CSS) (Hudgins et al. ; Nadeau et al. ), heterogeneous stocks (HS) (Hitzemann et al. ), and, more recently, Laboratory Strain Diversity Panels (LSDP) drawn from the Mouse Phenome Project (Paigen and Eppig ) for association studies.
Although the major use conceptualized for RI lines was linkage analysis (Bailey ), with the expanded sizes of many RI panels they are now being used to support analysis of more complex polygenic traits (Markel et al. ; Williams et al. ). Similarly, CSS and LSDP resources are being used for the genetic analysis of polygenic traits. CSS have a simplified genetic structure with only one chromosome differing between a single CSS and the parental recipient strain, a characteristic not shared with the other resources (Nadeau et al. ). The HS are significantly different than RI lines or CSS in that they typically contain multiple inbred strain progenitors, which potentially increases the level of genetic diversity represented in the resource (Yalcin et al. ). The LSDP were recently envisioned to adapt many of the whole-genome association technologies being developed by the human genetics community (Grupe et al. ; Bogue and Grubb ; Liao et al. ; Pletcher et al. ; McClurg et al. ; Payseur and Place ). In theory, the LSDP should encompass large amounts of variation, but in practice, since analyses of LSDP resources has largely been limited to panels of classical inbred strains, the diversity is most likely restricted to . Similar to LSDP resources, a more recently developed resource called the Collaborative Cross (CC) was designed to incorporate large amounts of variation (Threadgill et al. ; Churchill et al. ; Valdar et al. ). The CC is a mammalian genetic reference population that was designed to have controlled randomization of genetic factors, which is essential for causal inference. The CC was designed as a panel of recombinant inbred lines derived from eight parental inbred strains through a mating scheme that minimizes unpredictable genomic interactions between strains and optimizes the contribution from each parental strain. The selection of the parental strains was based upon historical breeding records and suspected relationships drawn from sparse maps of genetic variation.
Herein we sought to reanalyze the structure of genetic variation present in various mouse genetic resources using genome resequencing data (). We found that the vast majority of resources capture very small amounts of the existing variation and the variation that is captured is not randomly distributed. Unlike other resources, the CC has a high level of variation capture that is normally distributed across the genome. This structure is similar to that found in humans and other randomly breeding mammalian species, showing that the CC is an ideal model for systems biology analyses.
All genotype data used in this study were obtained from the National Institute of Environmental Health Science’s “Resequencing and SNP Discovery Project” (). These data contain over 109 million genotypes that identified 8.3 million SNPs spanning the 19 autosomes, the sex chromosomes, and the mitochondrial genome (). The 15 resequenced strains include 11 classical inbred strains (129S1/SvImJ, A/J, AKR/J, BALB/cBy, C3H/HeJ, DBA/2J, FVB/NJ, NOD/LtJ, BTBR
/J, KK/HlJ, and NZW/LacJ) and four wild-derived strains (WSB/EiJ, PWD/PhJ, CAST/EiJ, and MOLF/EiJ), representing the , , subspecies and , a subspecies that arose by natural hybridization between and (). In addition, the genotypes of the fully sequenced and annotated C57BL/6J genome were used. Incomplete genotypes were imputed as described previously (Roberts et al. ).
One example was chosen from each of the five major types of resources based on widespread or potential use. In all cases the example represented the maximal amount of diversity captured among similar resources. The BXD, derived from C57BL/6J and DBA/2J by B. Taylor, L. Silver, and R. Williams, was chosen as the prototypical RI line panel because of its past and current popularity (Taylor ; Peirce et al. ). The representative chromosome substitution strain panel was B.P generated by J. Forejt, which has PWD/Ph chromosomes introgressed into the C57BL/6J background. The Northport HS derived from A/J, AKR/J, BALBc/J, CBA/J, C3H/HeJ, C57BL/6J, DBA/2J, and LP/J was used as the example of heterogeneous stock (Hitzemann et al. ). The Collaborative Cross is an RI line panel produced from the eight parental inbred strains A/J, C57BL6/J, 129S1/SvImJ, NOD/LtJ, NZO/HlLtJ, CAST/EiJ, PWK/PhJ, and WSB/EiJ (Threadgill et al. ; Churchill et al. ). Finally, since the emergence of the Mouse Phenome Project (Paigen and Eppig ), several panels of inbred strains have been considered for association studies (Bogue and Grubb ; Liao et al. ; McClurg et al. ; Payseur and Place ). The LSDP described by Payseur and Place was used as a representative because it is composed only of classical inbred strains, including A/J, AKR/J, BALB/cByJ, BTBR T +tf/tf, BUB/BnJ, CBA/J, CE/J, C3H/HeJ, C57BL/6J, C57BLKS/J, C57L/J, C57BR/cdJ, C58/J, DBA/2J, FVB/NJ, I/LnJ, KK/HIJ, LP/J, MA/MyJ, NOD/LtJ, NON/LtJ, NZB/B1NJ, NZW/LacJ, PL/J, RIIIS/J, SEA/GnJ, SJL/J, SM/J, SWR/J, and 129S1/SvImJ. Other inbred panels that also include wild-derived strains have not been useful for association mapping because of the large number of private polymorphisms contributed by strains derived from other subspecies.
Estimates of the polymorphism diversity captured by each resource represent best-case scenarios since they assume all diversity present in the parental strains is captured by the derived resources. Genetic diversity can be estimated directly in the BXD RI and the B.P CSS because the parental strains have been sequenced. In the remaining resources it was necessary to substitute sequenced strains for those that have not been sequenced. These substitutions were based on genetic similarity estimated using genotypes at SNPs distributed along the entire genome (Petkov et al. ). Five of the parental strains in the Northport HS have been sequenced and include A/J, AKR/J, C3H/HeJ, C57BL/6J, and DBA/2J. The remaining three strains were substituted by a sister substrain (BALBc/J was substituted by BALB/cBy), a related strain (LP/J was substituted by BTBR
/J), or a Castle strain that will overestimate the diversity present in this panel (CBA/J was substituted by NZW/LacJ). Six of the parental strains in the Collaborative Cross have been sequenced: 129S1/SvImJ, A/J, C57BL/6J, NOD/LtJ, WSB/EiJ, and CAST/EiJ. The remaining two strains were substituted by strains from similar origins (NZO/HlLtJ was substituted by NZW/LacJ and PWK/PhJ was substituted by PWD/PhJ). Finally, for the LSDP we used all 12 classical inbred strains plus WSB/EiJ. Although the number of strains used for our analyses is significantly lower than in the original panel (Payseur and Place ), the WSB/EiJ strain is a larger contributor to the diversity than any single classical strain or group of classical inbred strains combined (Yang et al. ), suggesting that this will be an accurate representation of existing panels.
italic
fig
#text
The recent explosion in genetic variation data for mice made possible by the resequencing of 15 mouse inbred strains () allows us to accurately determine and compare the polymorphic architecture of different mouse genetic resources. The most widely used resources suffer from very low rates of polymorphism capture (all extant RI lines, RCS, and the B.A and B.129 CSS) or medium levels of polymorphism capture that is nonuniformly distributed (B.P CSS, B6.CAST CON, Northport, and Boulder HS, and the LSDP). Although the proportion of the genome being interrogated with these resources does not limit their use for discovering subsets of functional gene variants controlling specific phenotypes, it greatly impairs their utility for genome-wide systems biological analyses. In addition, differences in allele frequency among the resources impact the relative allele strength that can be detected, with a consequential effect on the number of functional gene variants that can be detected by a particular resource. The common ancestry, dominated by , of many the strains that have contributed to most mouse genetic resources has resulted in a dramatic reduction in the pool of available gene variants for genome-wide discovery and, more importantly, may complicate their use for systems-level analyses of mammalian biology that is dependent on high levels of uniformly distributed genetic variation.
The CC represents a resource that has optimal polymorphism architecture for system biological applications. In particular, the uniform distribution of the high level of variation captured is ideal to support global analysis of complex biological systems that is most efficiently achieved using experimental designs that employ multifactorial perturbations (Fisher ). Although the allele frequency distribution in the CC is not necessarily the best to detect the effects of any particular polymorphism, it is representative of natural populations and should outperform all resources for trait correlation analysis, which is the foundation of systems genetics, and all but the resources with only two parental strains in detection of specific gene functional variants. However, the resources with only two parental strains capture much lower levels of available polymorphisms, and the captured polymorphisms are not uniformly distributed, greatly reducing their genome-wide utility for systems biology applications. With the shift in complex trait gene discovery to humans that has been made possible by affordable high-density genotyping of large numbers of phenotyped individuals, the mouse will be taking a new role in biological research, that of a model to support mammalian systems biology investigations. Our analyses demonstrate that the CC represents a dramatic improvement over other genetic resources since it is the only resource that can serve this role based on the level, distribution, and allele frequency of captured polymorphisms. The overall performance of the CC is particularly remarkable given that the original choice of parental strains represented a compromise between the practical desire to take advantage of existing resources such as genome sequence, mapping panels, and ES cell lines and the ultimate goal of maximizing diversity (Churchill et al. ). |
Trisomy 21 (Ts21) is one of the most prevalent serious congenital malformations of genetic origin and the most common human aneuploidy compatible with survival. In the United States, 1 of every 733 live births has Ts21 (CDC ). Worldwide about 220,000 infants with Ts21 are born each year with phenotypes collectively referred to as Down syndrome (DS) (Christianson et al. ). Individuals with DS have subsets of approximately 80 clinical phenotypes, including cognitive impairment, craniofacial dysmorphology, congenital heart defects, gastrointestinal tract abnormalities, acute megakaryoblastic leukemia, immunologic defects, endocrine abnormalities, neuropathology leading to dementia, and dysmorphic physical features. To characterize the variability and origin of the many characteristic features of DS, multiple phenotypes have been studied during fetal and postnatal development (Delabar et al. ). The incidence and severity of specific DS phenotypes are influenced by genetic, environmental, and stochastic factors that occur throughout development and after birth (Cohen ; Epstein ).
The long arm of human chromosome 21 (Hsa21) contains 33.7 Mb and approximately 230 genes that are homologous to syntenic regions of mouse chromosomes 16, 17, and 10 (Fig. ) (Gardiner et al. ). The distal end of mouse chromosome 16 (Mmu16) contains 144 conserved and minimally conserved Hsa21 orthologs (Chromosome 21 gene function and pathway database, /) (Gardiner et al. ; Nikolaienko et al. ), and a number of segmental trisomy mouse models have been made with portions of this chromosomal region at dosage imbalance (Table ). The most widely used and well-studied mouse model of trisomy and DS phenotypes is the Ts65Dn (hereafter Ts65Dn). This segmental trisomy model has a small translocation chromosome comprising the distal region of Mmu16 attached to the centromeric end of Mmu17 (Davisson et al. ; Reeves et al. ) and contains about half of the Hsa21 gene orthologs (Hattori et al. ). Ts65Dn mice show DS-related phenotypes, including reduced birth weight, cognitive and behavioral impairments, craniofacial abnormalities, perinatal lethality, cardiovascular malformations, and neurologic structural deficiencies (Baxter et al. ; Belichenko et al. ; Cooper et al. ; Holtzman et al. ; Lorenzi and Reeves ; Moore ; Richtsmeier et al. ; Roper et al. ; Rueda et al. ). A number of phenotypes characterized in Ts65Dn mice have been used as a standard to compare the incidence and severity of trisomic phenotypes in other mouse models (Aldridge et al. ; Arron et al. ; Olson et al. , , ; Richtsmeier et al. ; Sago et al. ; Siarey et al. ). Other segmental trisomies of Mmu16 include Ts(12;16C-tel)1Cje and Dp(16Cbr1-ORF9)1Rhr (Ts1Cje and Ts1Rhr, respectively). Additional models can be made when a third copy of a gene or region is added to or subtracted from existing models. Both Ms1Cje/Ts65Dn and Ms1Rhr/Ts65Dn were produced by breeding the corresponding monosomy of newly developed Ts1Cje and Ts1Rhr trisomies to the existing Ts65Dn mouse (Olson et al. ; Sago et al. ). Ts[Rb(12.Ts1765Dn)]2Cje (Ts2Cje) mice were identified after a fortuitous translocation of the T65Dn marker chromosome (Villar et al. ). Owing to a conservation of genetic content and developmental processes in human and mouse, the mouse has served as an effective research model for many DS phenotypes.
Hsa21 genes have also been introduced into mouse cell lines to produce transchromosomic mice through microcell-mediated chromosome transfer and embryonic stem cell technology (O’Doherty et al. ; Shinohara et al. ). The recently developed Tc(Hsa21)1TybEmcf (Tc1) mouse has germline transmission of an almost intact Hsa21 (proximal and distal gaps omit approximately 10% of Hsa21 sequence and 8% of Hsa21 genes) but tissue-specific variability in cellular trisomy. Tc1 mice exhibit DS-like neurologic, behavioral, cardiovascular, and craniofacial abnormalities (O’Doherty et al. ).
Additional mouse models with three copies of segments of mouse chromosomes homologous to Hsa21 have been used to understand the gene-phenotype relationship of DS. These models contain dosage imbalance of some Hsa21 orthologs as well as large segments of nonhomologous genetic material. Trisomy 16 (Ts16) embryos are trisomic for all of Mmu16 (∼98 Mb), including the Hsa21 homologous region on the distal end of Mmu16, and also contain trisomic regions homologous to Hsa3, 8, 12, 6, and 22. Ts16 offspring die perinatally and this has limited studies in this model to developmental phenotypes, including edema and fetal cardiac, neurologic, and thymic abnormalities (Epstein et al. ; Hiltgen et al. ; Miyabara et al. ). The Ts43H (Ts43H) mouse is trisomic for 30 Mb of proximal Mmu17 and has been investigated for DS-related behavioral and gene expression phenotypes (Vacik et al. ). The breakpoint on the T43H segmental chromosome occurs in the Hsa21 homologous region and the Ts43H model is trisomic for at least nine Hsa21 orthologs found on Mmu17. Because of the large trisomic Mmu17 region, Ts43H has been estimated to have an overall greater number of genes at dosage imbalance than Ts65Dn or Tc1 mice. DS-like phenotypes identified in the Ts43H and Ts16 models could be due to three copies of Hsa21 orthologs, trisomy of large genomic regions, or trisomic heterogeneity that disturbs distinct pathways but leads to similar phenotypes.
Transgenic mouse models with single Hsa21 orthologous genes or regions at dosage imbalance have also been created (Altafaj et al. ; Ema et al. ; Kola and Hertzog ; Roubertoux et al. ; Smith et al. ). These models, along with mice that have a specific gene or region knocked out or deleted, provide valuable data for understanding gene function, especially when analyzed in parallel with segmental trisomy models. Because of differences in quantitative gene expression, absence of potential regulatory regions, and limitations of splice variants in transgenic mice, careful consideration must be given when analyzing these models. Although single trisomic genes may have a large effect on a specific phenotype, spatial and temporal gene expression must be accurately compared with segmental trisomy models as well as with individuals with DS.
The most useful intraspecies assessments compare well-defined phenotypes of trisomic and control euploid mice at similar ages or developmental stages, examine identical tissues, and utilize experimental procedures employing similar rigor and precise quantification (Sago et al. ). Because of genetic differences, it is likely that only a subset of DS-like phenotypes will be represented in a mouse model and each model must be thoroughly evaluated for phenotypes that recapitulate the human condition. Intraspecies comparisons must consider additional factors, including accurate gene expression, and different patterns of expression or function of human or mouse genes or regions in mouse models (Gardiner ; Reeves ). Intraspecies comparisons between strains should either use a similar genetic background (optimally using littermate euploid control animals) or provide methodologic means to adjust for differences in genetic background (Olson et al. ; Roper and Reeves ; Sago et al. ) since it has been shown that different backgrounds may affect trisomic phenotypes (Villar et al. ). Other inherent differences between trisomic strains include presence/absence of an extra centromere, potential trisomic maternal environment during development, and presence of trisomy in every cell (Box ). Even with similar rigor, dissimilarities in findings may be the result of experimental and methodologic differences or phenotypic variation (Insausti et al. ; Lorenzi and Reeves ; Richtsmeier et al. ).
The incidence and severity of phenotypes in individuals with DS is highly variable. The importance of heterotrisomy (inheritance of multiple nonhomologous alleles) in specific regions has been demonstrated to correlate with pathogenesis in trisomy (Baptista et al. ). Incidence and severity of traits also vary in DS mouse models. For example, mandibular traits of Ts65Dn were demonstrated to be more variable in Ts65Dn than euploid mice (Richtsmeier et al. ). Interacting loci from trisomic regions may be important in producing an equivalent DS-like phenotype. Attenuated phenotypes in different murine models when trisomic content is reduced is most likely the result of interacting factors in separate trisomic regions (Olson et al. , ; Sago et al. , . A single gene with major effect may be modified by other trisomic loci, and such interactions may represent genes defined as sufficient or necessary for a specific phenotype (Olson et al. , ; Salehi et al. ). Phenotypic variation also illustrates the possible importance of interacting loci and intervening nongenic sequences in determining phenotype (Antonarakis and Epstein ; Antonarakis et al. ).
Modifier loci from nontrisomic regions may also impact incidence and severity of phenotypes. Modifier loci such as and are not located on Hsa21 but have been shown to have importance in DS childhood leukemia and heart abnormalities, respectively (Maslen et al. ; Vyas and Crispino ). Nontrisomic modifier loci may also be of importance in mouse models since many trisomic models cannot be inbred and therefore are maintained on a mixed background (O’Doherty et al. ); Paz-Miguel et al. ; Vacik et al. ). The effect of genetic background on the cardiovascular development of Ts16 embryos indicates that frequency and timing of abnormalities in pharyngeal arch arteries vary based on the genetic background in the four strains tested (Villar et al. ). The importance of background and modifier loci needs to be investigated in future intraspecies studies.
Teaming comparative genomic analysis with developmental studies of DS models has the power to uncover the genesis of a specific phenotype by identifying where in development abnormal diverges from normal. Embryonic studies with Ts16 embryos provided the first insights into the mechanisms by which dosage imbalance of trisomic genes may impinge upon normal developmental processes in complex systems such as cardiovascular (Miyabara et al. ) and neurologic development (Ozand et al. ; Singer et al. ). Identifying the spatial, temporal, and molecular defects leading to an abnormal phenotype requires precise identification and quantification (e.g., cell number, volumetric analysis, complete histologic examination) since such differences may be small at their inception and may occur only in a subset of trisomic individuals. A phenotype-based analysis throughout development may not only lead to the identification of genes causing the phenotype, but may lead to identification of a particular pathway that may be important for therapeutic intervention. A phenotype-based intraspecies comparison may offer insight into the gene-phenotype interaction if development is different among trisomic models.
Developing trisomic humans and mice have been described as “developmentally delayed” (Holtzman et al. ; Reeves et al. ; Wisniewski ). In trisomic mouse models, this unfortunate nomenclature has often been used to describe developing and neonatal mice that are smaller than euploid littermates yet fails to define specific areas of dissimilar development. Ts16 embryos show reduced brain weight and disrupted neuronal transmitter system development (Ozand et al. ; Singer et al. ), reduced number and delayed maturation of thymocytes (Epstein et al. ), and reduced endocardial mesenchymal cells (Hiltgen et al. ) compared with euploid littermates. To date, little data reflect upon the delays associated with prenatal development in particular structures at specific embryonic stages in other trisomic models. It is unclear if all structures are delayed at the same developmental point or only those that later become dysmorphic. If there is a cellular paucity in a structure during development, is it only because of miscues in gene expression relating to those particular cells? Does a phenotypic change in a single population of cells or structure lead to abnormalities in other cells or structures? If a structure is “delayed,” does the window of opportunity for tissue interaction close and thus the derived tissues can never recover? Is each area of dysmorphology under separate genetic control? Do many developmental abnormalities combine to produce a significant phenotype? These questions remain to be answered by thorough investigation.
#text
Ts21 is the most common genetic cause of mental retardation and cognitive impairment is found in all individuals with DS. By adulthood the brain is microcephalic with proportionately greater reductions in the hippocampus, prefrontal cortex, and cerebellum (Epstein ; Pennington et al. ). Gross structures as well as cellular components have been quantified in mouse models to investigate the association between trisomic brain structure and function (Belichenko et al. ; Cooper et al. ; Holtzman et al. ; Kurt et al. ; Reeves et al. ). A comparison of the cerebellum and the cerebrum in different trisomic models has suggested interesting connections between pathology and dosage imbalance. Ts65Dn mice display a small cerebellum, with a reduction in cerebellar granule cell density that recapitulates the deficiency seen in individuals with DS (Baxter et al. ). A diminution in Purkinje cell density is also characteristic of the adult Ts65Dn brain. Though there is no apparent reduction in size of the cerebrum, the shape of the Ts65Dn cerebrum is different compared with that of euploid littermates (Aldridge et al. ). Ts1Cje mice have a similar reduction in cerebellar volume as do Ts65Dn mice, but they have only a slight decrease in cerebellar granule cell density and no significant change in Purkinje cell density (Olson et al. ). Ms1Cje/Ts65Dn mice do not display a significant reduction in cerebellar volume and have a significantly reduced cerebellar granule cell density but no reduction in Purkinje cell density. A direct comparison between Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn mice revealed that cerebellar size deficit may not be directly related to the paucity in granule cell density. The Ts1Rhr cerebellum is slightly smaller but is not as reduced as either the Ts65Dn or the Ts1Cje cerebellum, with a shape distinctly different from euploid littermates (Aldridge et al. ; Olson et al. ). Although cerebellar granule cell and Purkinje cell density is not significantly altered, both size and shape of the cerebrum differ between Ts1Rhr and euploid littermates (Aldridge et al. ). Limited studies of Tc1 brain structure showed a reduction in cerebellar volume and granule cell density when compared with those of euploid littermates O’Doherty et al. (). These analyses illustrate that specific brain phenotypes may be under different genetic control and that a similar pathology may result from heterogeneous sources or not be related to an obvious cellular deficiency. Different mouse models may be more useful for an in-depth study of brain phenotypes (e.g., Ts65Dn for Purkinje cell deficit and Ts1Rhr for cerebrum size deficit).
Long-term memory research has supported the notion of a hippocampal dysfunction related to DS (Pennington et al. ). The hippocampus is thought to be important in learning and memory, two parameters that are affected in individuals with DS. Total hippocampal volume is not changed in Ts65Dn mice, though reductions in dentate gyrus volume and granule cells have been documented (Insausti et al. ; Lorenzi and Reeves ; Rueda et al. ). Like Ts65Dn mice, Ts1Rhr total hippocampal volume is not decreased compared with that of euploid, although no quantification of Ts1Rhr-specific hippocampal regions and cellular constituents has been done (Olson et al. ). When survival and proliferation of Ts65Dn hippocampal granule cells were examined in adult mice, no significant difference was observed in either proliferation or survival of granule cells (Rueda et al. ). There were, however, a smaller number of proliferating cells in the dentate gyrus of older Ts65Dn mice. Fewer proliferating cells in adult trisomic brains coupled with the reduction of basal forebrain cholinergic neurons may lead to increased neurodegeneration of cells in the adult brain (Cooper et al. ). No reduction of volume of the dentate gyrus was seen in six-day-old Ts65Dn mice, but significantly fewer dentate gyrus granule cells and mitotic cells at six days were found when compared with euploid littermates. This result suggested that differences in neurogenesis of granule cells may contribute to the lower number of granule cells in the dentate gyrus of Ts65Dn mice (Lorenzi and Reeves ).
Changes in hippocampal structure may affect learning and memory as well as long-term synaptic plasticity (Galdzicki and Siarey ). Ts65Dn mice have been characterized as hyperactive and with deficits in learning and memory as defined by the Morris water maze (Escorihuela et al. ; Holtzman et al. ; Reeves et al. ). Ts65Dn mice generally decrease the time taken to locate both visible and hidden platforms in successive trials (nonspatial and spatial learning, respectively), although their improvement is significantly less than that of euploid littermates. In the probe trial test, Ts65Dn mice spend less time in the quadrant that contained the platform. Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn mice were generated to correlate the genetic contributions of different Hsa21 homologous regions with behavioral characteristics associated with DS (Sago et al. ). Although none of the strains were different from euploid controls in the visible platform test, Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn were all deficient in the hidden platform test compared with euploid controls. By comparison, Ts65Dn mice were the most, Ts1Cje were slightly less, and Ms1Cje/Ts65Dn were somewhat less impaired. Ts65Dn mice also spent less time in a specific quadrant test, whereas Ts1Cje mice were a little more like euploid littermates. In contrast with Ts65Dn, Ts1Cje mice were found to be hypoactive and Ms1Cje/Ts65Dn mice were not different than euploid mice in their activity. The most significant differences between Ts65Dn and Ts1Cje mice were observed in the reverse probe dwell and crossing tests examining cognitive flexibility. Overall, learning defects in Ts1Cje mice were similar to but slightly less than those seen in Ts65Dn mice. Ms1Cje/Ts65Dn mice showed little impairment in most tasks and deficits were less than those seen in Ts65Dn mice.
Similar results were seen when synaptic plasticity was compared in Ts65Dn and Ts1Cje strains. Long-term potentiation (LTP) and long-term depression (LTD) are physiologic measures associated with learning and memory (Antonarakis and Epstein ). Comparing Ts65Dn and Ts1Cje mice bred onto a similar background (but not littermates) showed differences between the two strains in hippocampal electrophysiology (Siarey et al. , , . Ts65Dn mice showed reduction of LTP in CA1 and dentate gyrus areas and increased LTD in CA1 (Kleschevnikov et al. ; Siarey et al. , . LTP was decreased and LTD was increased in Ts1Cje mice but the overall changes in long-term synaptic plasticity were less dramatic than those in Ts65Dn mice. The implications from this study were that the contribution from the trisomic segment common to Ts65Dn and Ts1Cje was significant for synaptic plasticity but other important interacting genes are part of the additional trisomic region in Ts65Dn mice.
Interestingly, Tc1 mice with an entirely different genetic background had a significantly reduced LTP in the dentate gyrus and a trend toward hyperactivity. These results reflect the robustness of these phenotypes even with different potential modifying loci, transchromosomal dissimilarities, including the human origin of the trisomy, and possible trisomic mosaicism (O’Doherty et al. ). Other tests examining learning and memory still need to be addressed with the Tc1 model (Reeves ). Ts1Rhr mice showed no deficiencies in either Morris water maze or synaptic plasticity when compared to euploid controls (Olson et al. ). Ts43H mice were not found to be hyperactive but had spatial learning defects in the Morris maze test comparable to those seen in Ts65Dn mice (Vacik et al. ). Hyperactivity differences in Ts43H may be due to necessary Hsa21 orthologs not included in the Ts43H mice and the spatial learning deficits may mean additional Hsa21 orthologs or trisomic heterogeneity may play an important role in these traits.
The structural and functional neuroanatomical studies reveal that the relationship between trisomic gene content and DS-related phenotypes may be unique to each phenotype. In general, models with less trisomic genetic content exhibit an attenuated phenotype, including cerebellar size, Morris swim maze, and synaptic plasticity. Analyses with Ts65Dn, Ts1Cje, Ts1Rhr, and Ms1Rhr/Ts65Dn mice have demonstrated that the DSCR concept is incorrect. The comparative studies between segmental trisomic mice illustrate that the interaction between genes and/or regions in two different areas may be important to both the incidence and the severity of structural and functional neurologic phenotypes, thus superseding the “one gene-one phenotype” hypothesis.
One of the most commonly associated phenotypes of Ts21 is the distinct craniofacial features seen in all individuals with DS. Using a sophisticated 3D skull analysis of Ts65Dn mice, Richtsmeier et al () showed that Ts65Dn mice exhibit craniofacial abnormalities that resemble those found in individuals with DS. Similar features included microcephaly, brachycephaly, small flattened face, reduced interorbital distance, and both a small maxilla and mandible. In Ts1Cje mice, more than 80% of the measurements were conserved between the two models and affected the same bones of the skull (Richtsmeier et al. ). Ts65Dn mice, however, had brachycephaly and bones were affected to a slightly greater degree. Both mouse models displayed a smaller mandible than euploid littermates, with the reduction in size specific to the coronoid and angular processes, and also had slight morphometric changes that are specific to each strain.
In contrast with Ts65Dn and Ts1Cje mice, Ts1Rhr mice, with the putative DSCR at dosage imbalance, had a larger skull and an overall rostrocaudal elongation of the skull compared with euploid littermates (Olson et al. ). The mandible of Ts1Rhr mice was larger and had a different shape than Ts65Dn and Ts1Cje mice, with differences concentrated in the condyle, inferior ramus, and incisive alveolar. Ms1Rhr/Ts65Dn mice, with all of the genes at dosage imbalance the same as Ts65Dn except in the DSCR, showed similar but attenuated effects on the skull as seen in Ts65Dn and Ts1Cje mice. From these results, the DSCR was shown to contain genes that were not sufficient and largely not necessary to cause DS-like craniofacial abnormalities.
Light microscopy showed no gross differences in the craniofacial structure of Tc1 mice (O’Doherty et al. ). Simple vector measurements showed no differences in the skull of Tc1 mice but indicated reduction in mandibular structure compared with euploid littermates. The mandibular differences specified trisomic effects in the coronoid and angular processes similar to Ts65Dn and Ts1Cje mice. Differences in methodologies measuring craniofacial structure may account for additional abnormalities not ascertained in Tc1 mice. Alternatively, genes important in craniofacial structure may not have the same expression in the transchromosomic model as found in segmental trisomy models.
At birth, Ts65Dn mice have differences in craniofacial structure in the anterior face, anterior and posterior neurocranium, palate, and mandible (Hill et al. ). Although some differences in dysmorphologies were identified between adult and newborn Ts65Dn mice, an analysis of postnatal growth patterns between trisomic and euploid mice showed that many of these early changes led to differences seen in adult Ts65Dn mice. It has also been hypothesized that slight craniofacial differences seen in Ts65Dn and Ts1Cje mice could be due to developmental differences between the strains (Richtsmeier et al. ). Similar developmental hypotheses could be extended to each different model, with differences in the dosage imbalance of certain genes or regions causing small developmental alterations of craniofacial structure.
The congenital heart defects (CHD) present in 50% of DS neonates include atrial, ventricular, and atrioventricular (AV) septal and valvar defects (Freeman et al. ; Wessels et al. ). These malformations are typically attributed to AV canal abnormalities with failure of proper endocardial cushion formation or fusion in the inner curvature of the heart tube of DS individuals. However, complex malformations also involving the outflow tract (e.g., tetralogy of Fallot) contribute to the cardiac phenotype in many DS cases (Freeman et al. ; Wessels et al. ). In a study of DS neonates undergoing both cardiac physical exam and echocardiography within a month of birth, 66% of this DS group had detectable cardiac anomalies. AVSD was identified in one third of the cases with abnormal echocardiographic findings, while tetralogy of Fallot was found in one fifth of these DS CHD cases (McElhinney et al. ). Aberrant formation or maintenance of the aortic arch arteries, manifested through persistent ductus arteriosus or aberrant right subclavian artery, also occur at a higher frequency in DS fetuses and neonates than in the general population (Chaoui et al. ; McElhinney et al. ). Studies of DS individuals with CHD, especially those with segmental trisomy of only part of Hsa21, have been used to investigate the molecular basis of the cardiac malformations. Several candidate genes mapped to Hsa21 that may contribute to a cardiac phenotype have been identified: (Davies et al. ), (Barlow et al. ), and (Rothermel et al. ). As yet, no one gene has been found to cause the complex, variable cardiac anomalies. The presence of an extra copy of one or more interacting genes from the distal region of Hsa21 may lead to disruption in the process of septation. Murine models in which only a subset of the candidate genes is triplicated may implicate or exclude several of the current candidate genes as the primary factors determining the predisposition to cardiac malformations.
The presence or absence of cardiovascular malformation has been carefully characterized in some but not all DS models. Cardiovascular abnormalities in Ts16 show complete penetrance, are apparent as septation occurs, and in some ways parallel the heart defects seen in DS, with more than half of the embryos displaying a common AV canal (Miyabara et al. ). The cardiovascular phenotype of Ts16 embryos also demonstrates the limitations of using whole chromosomal interspecies comparisons to identify the critical genes at dosage imbalance with analysis of complex structures derived from multiple tissue types. In analyzing the Ts16 cardiac phenotype, misalignment of the endocardial cushions, disruptions in neural crest, and loss of extracardiac mesoderm that typically contribute to septation are all posited to contribute to the observed atrioventricular, conotruncal, and atrial septal defects (Waller et al. ; Webb et al. ). Yet the right aortic arch and persistent truncus arteriosus identified in Ts16 embryos also resemble the cardiovascular phenotype of DiGeorge syndrome, a human syndrome associated with deletion of genetic material from Hsa22q11 (Waller et al. ). Murine genes orthologous to the DiGeorge region of Hsa22 are found on the proximal region of Mmu16 and are triplicated in Ts16, but not in trisomies containing only distal Mmu16 (Ts65Dn, Ts1Cje, and derived lines). Therefore, the severe cardiac phenotypes characterized in Ts16 may be viewed as a combination of the mechanisms that contribute to cardiac phenotypes in both DiGeorge and DS. Attempts to eliminate the cardiovascular phenotype of the Ts16 mice with reduction of one candidate gene () to diploid levels did not significantly alter the cardiac phenotype (Lange et al. ). Ts65Dn lacks the syntenic Hsa22 region of Mmu16, yet the cardiac abnormalities identified thus far in the segmental trisomy include right aortic arch and intracardiac septal defects (Moore ). The low frequency of gross cardiac anomalies, coupled with selective loss of trisomic neonates, prevented identification of the cardiac phenotype in the Ts65Dn mice. This DS model had been reported to lack any cardiac phenotype, so identification of a cardiovascular phenotype, albeit at a low rate, indicates the care that must be taken in characterization of each potential phenotype in each DS model. Though Ts65Dn mice do not have the severe phenotype of complete AV canal commonly associated with DS, the etiology of defects in the great vessels arise with abnormal formation and/or regression of the aortic arch arteries. Some aspects of DS CHD, such as tetralogy of Fallot, persistent ductus arteriosus, and aberrant right subclavian artery, also have their origins in aortic arch architecture. Therefore, shared elements of DS, Ts65Dn, and Ts16 cardiac phenotypes suggest some component of the aortic arch and outflow tract malformations may be attributed to the Hsa21 orthologs on distal Mmu16 .
The cardiac phenotype of the Tc1 mouse at embryonic day 14.5 (E14.5) resembled the abnormalities typically seen in DS CHD (O’Doherty et al. ). Seven of 11 Tc1 mice had an interventricular septal defect (one with overriding aorta), while one had unfused AV cushions. It should be noted that 20% of the euploid mice also had an interventricular septal defect and no later stages of development were presented, although the transmission rate in Tc1 on the F background was approximately 40%, lower than the 50% expected. Because failure of the cardiovascular system to form and function properly contributes to perinatal lethality in mouse and man, the non-Mendelian trisomic transmission rates seen in multiple DS models may indicate perinatal lethality of the most severely affected trisomic embryos due to cardiac or other anomalies.
Other researchers have begun to use the DS model mice as a primary line from which other transgenic lines may arise or be specifically created (Lange et al. ; Olson et al. ; Salehi et al. ; Villar et al. ). Definitive characterization of the cardiac phenotype in the Ts65Dn mice and other DS models as primary lines is crucial for ascertaining the effect of further genetic modifications produced by secondary modifications to these lines. The power of mouse models is that the genomic complement can be exactly manipulated and defined; we must therefore be just as exacting in identifying or, just as important, excluding the presence of DS phenotypes in the mice. Comparison of trisomic gene content and severity of cardiac phenotypes between Ts16, Ts65Dn, and other murine DS models may narrow the candidate regions of Mmu16, and hence Hsa21, responsible for different components (such as AV canal vs. outflow tract defects) of the complex and variable forms of DS CHD.
Powerful tools to analyze gene expression patterns have revealed a multitude of changes attributed to triplication of genes, yet results underscore the complexity of analyzing a multiphenotype, multigene syndrome such as DSIn humans, microarray analysis of fetal and adult tissue as well as cell lines derived from individuals with Ts21 show higher average gene expression from Hsa21 genes (FitzPatrick et al. ; Giannone et al. ; Mao et al. , . The dysregulation did not include all genes sampled from Hsa21, and secondary effects of increased transcript levels were noted on genes of nontrisomic chromosomes. Tissue-dependent patterns of overexpression were observed, further illustrating the complexity of correlating trisomic genes with transcript overexpression (Mao et al. ).
Gene expression studies in mouse models have shown similar complexity of results and interpretation. Although varied tissues and ages of mice were examined, analyses of neonatal and adult tissues of both Ts65Dn and Ts1Cje mice showed a near average 1.5-fold overexpression of triplicated genes (Amano et al. ; Dauphinot et al. ; Kahlem et al. ; Lyle et al. ; Saran et al. ). In addition, no average overexpression of genes found in two copies was seen either from nontriplicated regions of Mmu16 or the disomic chromosomes. Although average expression of trisomic genes was generally 1.5-fold, some genes in three copies were underexpressed, overexpressed to a greater degree, or unchanged. Differences in gene expression were specific to the tissue and developmental stage of the sample. Significant dysregulation of gene expression in disomic genes was reported in a number of studies, including one study that illustrated a global secondary disruption of gene expression due to trisomy (Saran et al. ). In Tc1 E14.5 embryos, microarray analysis on human arrays showed 39 of 131 Hsa21 genes and only 9 of 22,078 non-Hsa21 probe sets were overexpressed compared with euploid littermates (O’Doherty et al. ). When gene expression was examined in Ts43H mice, 20 brain-specific genes at dosage imbalance gave an average of 1.2-fold increased expression of euploid, with expression of only two genes reaching 1.5-fold expression (Vacik et al. ). In addition, 12 genes on the nontrisomic portion of chromosome 17 had expression levels that were 90% of euploid level. Brains from Ts2Cje mice exhibited a 1.5-fold expression level of specific trisomic genes comparable to Ts65Dn and different from euploid. Further data and analyses in both humans and mice are needed to reach biologically significant conclusions (Antonarakis and Epstein ; Reeves ).
A phenotype-based intraspecies comparison of mice will help to elucidate how Ts21 leads to DS phenotypes. Three recent examples that examined development of DS-like phenotypes in mouse models implicated particular genes and pathways that may be important in specific phenotypes. These studies also illustrate the complexity of the gene-phenotype relationship in DS.
Previous observations in Ts65Dn mice showed an age-related atrophy and loss of basal forebrain cholinergic neurons (BFCNs) in the medial septal nucleus (Holtzman et al. ). In Ts65Dn mice, although nerve growth factor (NGF) levels were greater than normal, NGF retrograde transport was severely reduced. Normal size and number of BFCNs were found after delivering NGF directly to the BFCN cell bodies (Cooper et al. ). Retrograde transport of NGF in Ts1Cje mice was about 70% of control and significantly greater than Ts65Dn mice, but no significant differences in size and number of BFCNs nor the abnormal axonal phenotype were observed in Ts1Cje mice (Salehi et al. ). Protein levels of full-length , triplicated in Ts65Dn but not in Ts1Cje mice, were linked to the abnormal retrograde transport of NGF. NGF transport, however, was not completely returned to normal in either Ts1Cje or Ts65Dn mice with only two functioning copies of . From these experiments it was concluded that abnormal dosage of combined with the trisomy of other regions was an important factor in the deficient transport of NGF and cholinergic neurodegeneration.
Mice with defects in the NFAT signaling pathway display many phenotypic similarities to DS, including neurologic, craniofacial, and endocardial cushion abnormalities (Arron et al. ). Though not all phenotypes occur in every model, -/-; -/- double knockout mice display aspects of brachycephaly, midface hypoplasia, and dysmorphic mandible. , an inhibitor of calcineurin/NFATc signaling, triplicated in Ts65Dn and Ts1Cje mice, and expressed in higher levels in DS fetuses, was selected as a candidate gene for craniofacial defects. was also selected as a candidate gene and was shown to regulate the calcineurin/NFAT signaling pathway in response to fibroblast growth factor 8 (FGF8). Dyrk1a and Dscr1 were shown to synergistically block NFAT-dependent transcription. Transgenic overexpression of Dyrk1a alone and with Dscr1 led to vascular defects and a failure in heart valve elongation, respectively. Interestingly, cortical neurons of Ts1Cje E13.5 embryos showed an increase of expression but whole heads of E11.5 and postnatal day 1 (P1) hippocampal neurons did not have increased Dyrk1a or Dscr1 protein levels or alterations in phosphorylation of NFATc. The authors conclude that during brief developmental periods an increased dosage of and reduces NFAT transcriptional activity and leads to mild versions of NFATc phenotypes.
The above example illustrates the complexity of intraspecies comparison in DS mouse models. Ts1Rhr mice (trisomic for but not ) display craniofacial defects that are distinctly different from Ts65Dn and Ts1Cje mice (Olson et al. ). Ms1Rhr/Ts65Dn mice (trisomic for all genes in Ts65Dn except those found in the “DSCR,” including ) exhibit slightly attenuated craniofacial abnormalities compared to Ts65Dn and Ts1Cje and do not have brachycephaly. Furthermore, Tc1 mice have only two copies of and mandibular abnormalities similar to Ts65Dn mice. A meta-analysis of all strains on a similar genetic background, at similar developmental timepoints, using stringent methodologic analyses will be useful to understand the complete role of and in craniofacial structural abnormalities. Similar arguments could be made for the role of these genes in heart defects because of a recent report finding heart defects in Ts65Dn mice (Moore ). Dysregulation of multiple pathways may lead to similar DS-like phentoypes and it will be important to understand which pathways are important in Ts21.
Ts65Dn and Ts1Cje mice were used to examine the origin of the cerebellar size deficit and paucity of granule cells (Baxter et al. ; Olson et al. ). Reduction in size of the Ts65Dn cerebellum was observed throughout development until and including P6 (Roper et al. ). A granule cell deficiency was seen throughout development and traced to a deficit in mitosis of granule cell precursors at the day of birth. The mitotic deficit was linked to a decreased response by trisomic granule cell precursors to sonic hedgehog (Shh), a molecule important in proliferation of granule cell precursors. Treatment of newborn mice with a Shh pathway agonist overcame deficits in mitosis and the number of granule cell precursors six days later. Though the pathogenesis of the cerebellar deficit was described and linked to a cellular mechanism, the trisomic genetic mechanism leading to the cellular deficit is still unknown. Though is not found on Hsa21, it is possible that it may be linked to many DS phenotypes (Roper et al. ).
In these examples a single gene or pathway may provide a major factor in the development of a DS-like phenotype. As noted from the microarray studies, triplication of genes and/or genomic regions may lead to dysregulation of disomic genes and a number of different pathways that may appear unrelated to the initial trisomic insult. A number of distinct pathways, however, can produce a single phenotype or be used to correct a phenotype; therefore, therapy derived for a particular phenotype may involve a pathway that may be only indirectly disturbed by trisomy. Correlation of gene misexpression with trisomic phenotypes is among the next challenges in understanding DS phenotypes.
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
e
s
o
f
D
S
m
o
u
s
e
m
o
d
e
l
s
p
r
o
v
i
d
e
a
p
o
w
e
r
f
u
l
t
o
o
l
t
o
i
d
e
n
t
i
f
y
g
e
n
e
s
t
h
a
t
a
r
e
m
i
s
r
e
g
u
l
a
t
e
d
b
y
t
r
i
s
o
m
y
,
p
l
a
c
e
t
h
e
s
e
g
e
n
e
s
i
n
p
a
t
h
w
a
y
s
l
e
a
d
i
n
g
t
o
p
h
e
n
o
t
y
p
i
c
a
b
n
o
r
m
a
l
i
t
i
e
s
,
a
n
d
u
n
d
e
r
s
t
a
n
d
r
e
g
u
l
a
t
o
r
y
n
e
t
w
o
r
k
s
,
i
n
c
l
u
d
i
n
g
c
o
m
p
e
n
s
a
t
i
o
n
a
n
d
i
n
t
e
r
a
c
t
i
o
n
.
I
n
t
h
e
f
u
t
u
r
e
,
a
d
d
i
t
i
o
n
a
l
m
o
d
e
l
s
n
e
e
d
t
o
b
e
d
e
v
e
l
o
p
e
d
w
i
t
h
t
h
r
e
e
c
o
p
i
e
s
o
f
H
s
a
2
1
h
o
m
o
l
o
g
o
u
s
r
e
g
i
o
n
s
n
o
t
f
o
u
n
d
i
n
c
u
r
r
e
n
t
m
o
d
e
l
s
.
E
x
i
s
t
i
n
g
m
o
d
e
l
s
c
a
n
b
e
m
o
d
i
f
i
e
d
b
y
a
d
d
i
n
g
o
r
s
u
b
t
r
a
c
t
i
n
g
g
e
n
e
s
o
r
r
e
g
i
o
n
s
t
o
i
s
o
l
a
t
e
s
p
e
c
i
f
i
c
g
e
n
e
-
p
h
e
n
o
t
y
p
e
r
e
l
a
t
i
o
n
s
h
i
p
s
.
I
n
f
u
t
u
r
e
c
o
m
p
a
r
a
t
i
v
e
s
t
u
d
i
e
s
,
i
t
i
s
i
m
p
o
r
t
a
n
t
t
o
e
x
a
m
i
n
e
t
i
s
s
u
e
s
a
t
m
u
l
t
i
p
l
e
t
i
m
e
p
o
i
n
t
s
d
u
r
i
n
g
d
e
v
e
l
o
p
m
e
n
t
a
n
d
t
o
u
s
e
i
d
e
n
t
i
c
a
l
a
n
d
r
o
b
u
s
t
m
e
t
h
o
d
o
l
o
g
i
e
s
.
I
n
t
r
a
s
p
e
c
i
e
s
p
h
e
n
o
t
y
p
e
-
b
a
s
e
d
a
n
a
l
y
s
e
s
s
h
o
w
p
r
o
m
i
s
e
i
n
f
i
n
d
i
n
g
w
a
y
s
t
o
d
e
f
i
n
e
p
h
e
n
o
t
y
p
i
c
e
t
i
o
l
o
g
y
a
n
d
t
o
a
m
e
l
i
o
r
a
t
e
o
r
p
r
e
v
e
n
t
s
p
e
c
i
f
i
c
D
S
p
h
e
n
o
t
y
p
e
s
. |
TRP cation channels play an important role in signal transduction of numerous organisms. For example, in , TRP channels are known to mediate the response to light (; ; ; ; ) and in mammals they are involved in detections of changes in the environment. TRP channels gate in response to a myriad of stimuli, including cold or hot temperatures, natural chemical compounds (menthol, camphor, and capsaicin), and mechanical stimuli. These channels are crucially involved in physiological processes, e.g., photoreception, pheromone sensing, taste perception, thermosensation, pain perception, mechanosensation, perception of pungent compounds (mustard, garlic), renal Ca/Mg handling, smooth muscle tone, and blood pressure regulation (; ; , ; ; ; ; ). There are seven TRP subfamilies: TRPC, which includes the founding member of this family, the TRP (), TRPM, TRPV, TRPN, TRPA, TRPP, and TRPML (; ; ). In spite of the vast interest in these channels, the mechanism underlying stimulus-dependent regulation and gating of TRP channels remains largely unknown ().
Recent studies conducted in heterologous expression systems implied that members of the TRPV and TRPM subfamilies are voltage-gated channels, showing a temperature- and agonist-dependent shift of their voltage dependence (; ). It is, however, unclear whether other subfamilies of TRP channels share the same voltage-dependent gating mechanism ().
The photoreceptor cell is a powerful experimental system, currently being one of the few systems in which two members of the TRPC family, TRP and TRPL, are accessible to whole cell recordings and their physiological functions are well defined in vivo (; ; ; ; ). However, the photoreceptor cells are not accessible to single channel recordings of the native TRP and TRPL channels (; ). Often, channels expressed in heterologous systems differ in their properties from in vivo channels (). This, however, is not the case for the expressed TRPL channels in S2 cells, as they show similar properties to the native channel (). Therefore, results obtained in S2 cells expressing TRPL channels, which allows single channel analysis (see below), can be verified and provide physiological insight to the native system.
In the present study we explored the voltage dependence of expressed TRPL channels and show that it is not an intrinsic property of the channels, but arises from voltage-dependent Ca open channel block. A mathematical model constructed from experimental observations describing channel–Ca interaction indicated that it is the dissociation of Ca from the open channel that underlies the voltage dependence of the channel. Whole cell recordings from a mutant expressing the native TRPL channels indicated that Ca block also accounts for the voltage dependence of the native TRPL channels in the photoreceptor cells.
White-eyed null mutants were used.
A xenon high-pressure lamp (PTI, LPS 220, operating at 50 W) was used, and the light stimuli were delivered to the ommatidia by means of epi-illumination via an objective lens (in situ). The intensity of the orange light (Schott OG 590 edge filter) at the specimen, with no intervening neutral density filters, was 13 mW/cm.
Schneider S2 cells were grown in 25-cm flasks at 25°C in Schneider medium (Beit Haemek Biological Industries) supplemented with 10% FBS and 1% pen-strep. In the present study we used stably expressed TRPL-eGFP channels. Table S1 (available at ) show that there is no significant difference in the major properties of the TRPL- and TRPL-eGFP–expressed channels used in the current study.
For S2 cells, cells were seeded on polylysine-coated plates at a confluence of 25%, 24–72 h before the experiment. 24 h before the experiment, 500 μM CuSO was added to the medium to induce expression of the TRPL-eGFP channel. For ommatidia, dissociated ommatidia were prepared from newly emerged flies (<1 h after eclosion) and whole-cell, patch-clamp recordings were performed as previously described (). Whole-cell and single-channel currents are recorded at room temperature using borosilicate patch pipettes of 5–8 MΩ and an Axopatch 1D (Axon Instruments, Inc.) voltage-clamp amplifier. For single channel recordings, pipettes were coated with Dow Corning sylgard. For S2 cells, voltage-clamp pulses were generated and data were captured using a Digidata 1322A interfaced to a computer running the pClamp 9.2 software (Axon Instruments, Inc.). For ommatidia, Digidata 1200 and pClamp 8.0 software (Axon Instruments, Inc.) were used. Currents were filtered using the 8-pole low pass Bessel filter of the patch-clamp amplifier at 10 kHz and sampled at 50 kHz (single channel recordings) or filtered at 5 kHz and sampled at 20 kHz (whole cell recordings). To measure I-V curves with minimal distortions, only cells with low (<10 MΩ) series resistance were used and the series resistance was compensated to ∼80%.
The details of the solutions are presented in and . All solutions were titrated to pH 7.15. Free Ca was estimated using MaxChelator v. 2.50 () using tables cmc0204e.tmc and the following parameter settings: t = 25°C, pH = 7.15, I = 163.
Data is analyzed and plotted using pClamp 9.2 software (Axon Instruments, Inc.) and Sigma Plot 8.02 software (Systat Software, Inc.). For kinetics analyses, the probability of single channel openings (if there are actually two channels present) was estimated according to the equation given by .
where is the number of single openings. Single-channel events were identified on the basis of the half-amplitude threshold- crossing criteria. Histograms of open and closed durations were constructed conventionally as distributions of the logarithm of duration (in ms, 15 bins per decade) with exponential components fitted by the method of maximum likelihood (). The frequency density and fitted functions were plotted after a square root transformation. The number of probability density function components was determined using the automatic “compare models” routine of the pClamp 9.2 software at the confidence level of 0.99. Current amplitude histograms are fitted with Gaussian functions. Burst analysis was performed using the pClamp 9.2 automatic search with the maximum interval method of 1- and 5-ms intervals between bursts. For event histograms, only events longer than twice T were used (). The event histogram, which was constructed by this method, does not represent the channel open probability but the number of events.
We used a stochastic model () to simulate single channel activity and a differential model to simulate whole cell activity. Both models were solved using MATLAB 6.5 software.
The supplemental material (available at ) includes two sections. The first section explains how the rate constants of the model were derived experimentally. The second section contains three model insights: an explanation of why the short blocked state cannot be observed experimentally; demonstration of burst stabilization at positive membrane potential by the blocked state; and detailed explanation of why the dissociation of divalent ions from the blocked state is voltage dependent. In addition, the supplemental material includes three tables and a figure (Tables S1–S3 and Fig. S1): a comparison between expressed TRPL and TRPL-GFP; actual data used to derive the above rate constants; the parameters of the voltage-dependent divalent block model; and a figure demonstrating the above three model insights.
To explore, for the first time, the voltage dependence of light-activated conductance at the single channel level, we used TRPL channels expressed in S2 cells. This is because the native TRP and TRPL channels are not accessible to single channel recordings (; ) and heterologous expression of functional TRP channels has proven unsuccessful ().
Cell-attached patch clamp recordings from S2 cells expressing TRPL, at various membrane voltages in the presence of divalent ions, revealed a large increase in channel activity with the increase in membrane potential (). Consistent with the cell-attached data, whole cell recordings showed that cells expressing TRPL displayed outwardly rectifying I-V curves (using voltage ramps from −150 to 150 mV in 1 s) when exposed to extracellular solutions containing either Ca or Ba ions (, green and black curves, respectively). Removal of divalent cations from the extracellular solution resulted in an increase of inward current but did not cause a full linearization of the I-V curve (pink curve). When both intracellular and extracellular solutions were divalent free, the channels exhibited a linearization of the I-V curve (blue curve, = 11). The outward rectification of the I-V curve could be restored upon application of external Ca or Ba (, green and black curves, respectively). It can be argued that in the absence of divalent cations, the membrane becomes fragile and gives rise to a large leak current, which underlies the linear I-V curve. This possibility was ruled out by the following observations. First, shows that the divalent free solution affected only the currents at the negative membrane potentials (from −200 pA to approximately −2,000 pA), while at positive membrane potentials the currents remained virtually the same. Second, the reversal potential during divalent-free solution was ∼15 mV and not zero as expected from leak current, consistent with the reversal potential of intracellular Cs solution and extracellular Na solution. Third, after establishing a linear I-V curve, we applied either Ca- or Ba-containing solutions, and the outward rectification was reestablished ().
To further quantitate the blocking effect of Ca, we measured the affinity of Ca to the TRPL channel (). The affinity was calculated by measuring the degree of Ca block on the inward current during whole cell recordings using Ca-EGTA and Ca-EDTA buffer solutions, and a Kd of 163 ± 5 nM was measured (). The high affinity of TRPL to Ca and the steep dependence of the Ca block on TRPL channel activity have important physiological implications (see Discussion).
The results of indicate that the voltage dependence of the expressed TRPL channel is not an intrinsic property but arises from a high-affinity Ca block, which is relieved at positive membrane voltage.
To unravel the mechanism underlying the voltage- dependent I-V relationship of expressed TRPL channels and its dependence on Ca, we employed single channel measurements. Excised inside-out patches with pipettes containing divalent-free solution revealed that patches exposed to divalent-containing solution at the intracellular side showed no channel activity at the negative membrane potential and a significant activity at the positive membrane potential (, left). Membrane patches exposed to divalent-free solution showed similar channel activity at both negative and positive membrane potentials (, right). Ca exerted a more profound inhibition from the intracellular side relative to the extracellular side. This conclusion was derived from the complete inhibition of channel activity even at positive membrane potentials in excised inside out patches when exposed to 1.5 mM Ca (), while significant current was still observed in whole cell recordings in the presence of the same Ca concentration in the extracellular solution.
Several mechanisms could account for the Ca-dependent outward rectification of the TRPL channel. To determine if divalent cations could affect single channel conductance in a voltage-dependent manner, we examined the effect of membrane potential on various channel parameters in the presence of divalent cations. To measure the single channel conductance, we applied voltage ramps (from −150 to 150 mV in 1 s) in an excised inside out membrane patch that contained only few channels. The low frequency of channel activity at negative membrane potentials () made it difficult to obtain an I-V curve of single channel open conductance at negative membrane potentials. To overcome this difficulty we used superimposed current traces during voltage ramps from the same cell. In some cells after application of many voltag ramps, the frequency of channel openings increased spontaneously, showing many channel openings at negative membrane potentials, while at positive membrane potentials, the channel was virtually constantly open and relatively small closing events were observed. presents an example of such recording. The straight line is a linear regression curve fitting only the long channel openings. This curve also fits superimposed current traces of voltage ramps from the same cell before the enhanced channel activity was developed. A linear I-V curve of the open state of the channel was observed (, gray line). A single channel conductance of 110 pS ( = 5) was measured ( inset), which is similar to that obtained in a previous systematic study (), but differs from earlier reports by about a factor of 2 (). This discrepancy most likely arose from the different method used to calculate single channel conductance in the latter study. mainly used noise analysis from whole cell recordings, which usually gives a rough estimate of the actual single channel conductance. Indeed, several studies in which the single channel conductance of the expressed TRPL channel was determined directly from single channel analysis revealed single channel conductance very similar to that obtained in the present study, although different expression cell systems were used (i.e., COS, HEK, and Sf9 cells; ; ; ).
thus demonstrates that the open channel behaves as an ohmic resistor, ruling out the possibility of changes in single channel conductance as the source of the voltage dependence.
To further examine whether the outward rectification is due to the existence of several levels of channel conductance, we constructed event histograms (see Materials and methods) for the entire range of membrane potentials. shows two examples from positive (top) and negative (bottom) membrane potentials. A single Gaussian produced a best fit for the current distribution, indicating that the TRPL channel has only one channel conductance.
To address the possibility that the dependence of channel open probability, P, on membrane potential underlies the voltage dependence of the channel, we measured P directly as a function of membrane potential. shows that P depends on membrane potential. This conclusion is reached irrespective of the method used to calculate P (single channel patch clamp or whole cell recordings; , filled circles and open circles, respectively). The voltage dependency of P could result from the effect of voltage on the number of channel openings (opening frequency) or from the effect on the mean open time. The results of revealed that voltage affected the opening frequency. In particular, the channel opening frequency increases as depolarization increases (the number of channel openings at −150 mV is only 5% of that at 150 mV). The observed frequency of channel openings measured from single channel recordings correlates well with the channel open probability measured from whole cell recordings (, inset) and the two different measurements fit a curve (, solid line) derived from model calculations of divalent block (for details on the model see below and ). This close fit suggested that voltage dependence of the opening frequency is sufficient to account for the voltage dependence of the open probability.
We next examined whether the mean open time is also voltage dependent. An open time histogram is presented in two ways: in linear (, inset) and log scales (). The histogram in log scale is more revealing () and could be fitted by two exponents of a short time constant (0.1 ± 0.01 ms) and a longer time constant (0.57 ± 0.02 ms; = 11). The calculated values of the mean open time are plotted in as a function of membrane potential. In contrast to the frequency of channel openings (), the two time constants of channel closing, which determine the mean open time, show no voltage dependence. Thus, the frequency of channel openings is the sole parameter that determines the voltage dependence of the TRPL channel.
The data presented in revealed a strong inhibitory effect of Ca on TRPL channel openings. Therefore, we further examined the mechanism underlying this Ca-dependent inhibition and its relationship to the apparent voltage dependence of the channel ( and ). The observation that two exponents are required to best fit the open time histogram () suggests that two open states of long and short open time exist, though only a single conductance characterizes both (). The mean open time of both open states is voltage independent (). Given the strong inhibition of Ca on channel openings, () we investigated whether the mean open time is Ca dependent. Importantly, the open time histogram () exhibited two time constants, where the fast one was not sensitive to divalent cations (, left), while the slow one was ∼2.5-fold faster in the presence of divalent cations (, right). The results of further indicate that only the long open state is susceptible to open channel block, presumably because the short open state is too short to allow a Ca block (see model insights in the online supplemental material, available at ). Together, the data of –
indicate that Ca blocks the open channel because it affects the duration of channel openings (open channel block). In addition, depolarization relieves the channel from the Ca block, a typical characteristic of open channel block mechanism.
To better understand and characterize the properties of the TRPL channel in general and the nature of the Ca block in particular, we constructed a minimal kinetic model of the channel states (). In constructing the model we were guided by the previously shown experimental results. Specifically, may suggest that the transition from closed to open state is voltage dependent. However, the linear I-V curve at divalent-free solution () and the observation that divalent cations affect the mean open time () negates this possibility.
The TRPL channel revealed bursting behavior (). This feature is a prominent characteristic of TRP channels in general (e.g., ). Bursting behavior of a channel is known to arise from fast transitions of the channel between a conducting (open state) and nonconducting closed or blocked states (see model in ). The observation that two exponents are needed to best fit the open time histogram () already suggested that two open states of long and short open times exist, though only a single conductance characterizes both (). To show that the two open states also exist in Ca-free solutions where block is not expected, we constructed an open time histogram under these conditions, and two exponents were also needed to fit best the histogram (). At least three closed states of the channel must be assumed to account for the bursting behavior of the channel, in which two distinct burst types were observed that differed in the mean open () and closed times within the bursts ().
The model calculations tested the main issues of this work, namely the mechanism by which Ca blocks the channel. To accommodate the model to the result showing that the mean open time is strongly reduced by Ca () we assumed in the model calculations that Ca blocked the open channel (open channel block) and that depolarization relieved the channel from that block. Given that the TRPL channel has two open states, the results of suggested that only the long open state was susceptible to open channel block. Accordingly, a blocked state (B) was introduced into the model after the long open state (O). Since it was reasonable to assume that two open states with the same conductance share the same blocking properties, we added a blocked state (B) to the short open state (O) as well (see model insights below). Since demonstrated that the divalent ion block occurs from both sides of the membrane (see also ) we set no limitation on the direction of divalent cation entry to the channel.
The result of and clearly showed that the TRPL channel exhibits voltage dependence in the presence of divalent cations. This result suggests that either entry of the channel or its exit from the blocked state is voltage dependent. To distinguish between these two possibilities we measured the mean closed time within the burst as a function of membrane voltage (). shows that the mean closed time becomes shorter at depolarizing voltage, indicating that k (the rate constant of exit from the blocked state) is voltage dependent. This conclusion was strongly supported by the model calculations as explained in model insights (supplemental text and Fig. S1). The model parameters that are listed in Table S3 were derived from direct experimental measurements (supplement) and account for all the above arguments.
A powerful way to explore the properties of the TRPL channel in general and the nature of the Ca block in particular is to analyze the channel bursting behavior and its apparent voltage and Ca dependence. Bursts are fast transitions of the channel between the open and the adjacent closed states. Thus, parameters that affect the burst, affect the open state. We therefore investigated burst properties with and without divalent cations at various membrane potentials. The design of these experiments was based on model predictions. Importantly, burst duration and the number of openings within the burst were independent of membrane potential in divalent free conditions when no block was expected, as predicted by the model (, right). In contrast, in the presence of divalent cations, an increase in membrane potential markedly increased the number of openings within the burst and consequently burst duration (, left). Furthermore, at negative membrane potential there were fewer events in the burst in the presence of divalent cations, and the burst duration became shorter relative to its duration in divalent-free condition, indicating that the blocked state terminates the burst. We consistently found that the criterion for burst definition (i.e., the time interval between channel openings within the burst) did not affect the conclusion that burst duration is strongly dependent on membrane potential in the presence of divalent cations (unpublished data). In the presence of divalent cations at positive membrane voltage, the number of events in the burst increased relative to divalent-free conditions (), indicating a stabilizing effect of the divalent cations on the burst. This observation indicates that the rate constant of TRPL-Ca dissociation is voltage dependent (see Fig. S1).
Although the data of were not employed to construct the model, channel activity and bursting behavior, calculated by a stochastic version of the model fit well the experimental data (not depicted). Furthermore, the model correctly predicted the dependence of channel opening frequency on membrane potential (solid line in ). The differential version of the model also accurately predicted the outward rectifying I-V curve in the presence of divalent cations and the linear I-V relationship in their absence ().
Taken together, the excellent fit between the experimental data and model calculations strongly supports the important model prediction that TRPL–Ca dissociation (and not association) is the voltage-dependent step that underlies the voltage dependence of the TRPL channel.
The results of –
demonstrated that removal of Ca block by positive voltage activated the TRPL channels by increasing the frequency of channel openings. Furthermore, the predictions of the kinetic model and TRPL channel bursting behavior indicated that voltage-dependent increase of the dissociation rate constant of Ca from the blocked state underlies TRPL openings at positive membrane potential. Can open channel block by divalent cations, found in the expressed TRPL channels, also account for the voltage dependence of the native TRPL of the photoreceptor cells?
Before answering this question it is useful to outline several properties of the native versus the expressed TRPL channel and to address some of the problems involved in studying the effects of Ca on the channels in the native photoreceptor cell. In the native system, the TRPL channel can be studied separately without the effects of the highly Ca-permeable TRP channel, by using the null mutant, which expresses only the TRPL channels (). Although the characteristics of the native and expressed TRPL channel are similar, activation of the channel in the two systems is very different: Heterologously expressed TRPL channels in S2 cells are constitutively active (), while the native channels show their properties only during activation by light () or by metabolic stress (; ; see below). Moreover, it is difficult to isolate the effects of Ca on the native TRPL from its effects on the phospholipase Cβ (PLC), which activates TRPL, and its activity is strongly Ca dependent (; ). In wild-type fly, the major fraction of Ca influx is via the Ca-permeable TRP channel. The mutant, which expresses only TRPL channels, responds to sustained light transiently, revealing response inactivation due to an insufficient Ca influx for normal PLC activity during prolonged lights (). This response inactivation is the hallmark of the mutant phenotype ().
Long incubation of the photoreceptor cells in Ca-free medium is required to obtain total removal of Ca ions from the cytosol of these cells, due to their highly compartmentalized structure. However, prolonged Ca deprivation abolishes light activation of the native channels (; see also ). Therefore, we could examine the effect of Ca-free solution for limited times only. We examined the effect of reducing Ca levels on the I-V curve of the native TRPL channels either during shorts illuminations (1 s), or after metabolic exhaustion. Metabolic exhaustion induced TRPL channel openings in the dark after removal of ATP and GTP from the recording pipette. In both methods of channel activation, the magnitude of TRPL channel openings declined slowly (∼1 min) with time, and therefore, I-V curves were measured at maximal channel activity. When the extracellular solution contained 1.5 mM Ca, virtually no inward current was observed at negative membrane potentials (, blue curves). In contrast, when no divalent cations were added to the extracellular solution (Low divalent), ∼1–2 nA inward current develops at −60 mV (, black). Further reduction of the Ca level by adding EDTA (5 mM) to the intracellular and extracellular solutions resulted in >5 nA inward currents at −60 mV (, red curves). Importantly, the I-V curves measured after metabolic stress () or light activation () revealed similar dependence on Ca concentrations, indicating that Ca blocks the TRPL channel independent of the method of TRPL activation.
The large increase of the current in divalent-free conditions might arise from an unspecific leak current. This possibility was ruled out by the following observations. First, after the cessation of the light stimulus in divalent-free conditions, the dramatic increase of both inward and outward currents during light was strongly suppressed, returning to the original small leak current in the dark (, green). Second, channel activation by metabolic stress is characterized by a transient increase in conductance (). In divalent-free conditions after a nearly linear I-V curve was observed, the activated channels closed spontaneously and the current was reduced to a relatively small leak current (unpublished data). Together, these observations indicate that the near linearization of the I-V curve at divalent-free conditions is not the result of an increase in nonspecific leakage through the plasma membrane but a specific effect on the TRPL channels.
To quantitate the Ca-dependent block of the inward current relative to the magnitude of the outward current, we calculated the ratio between the inward and outward currents at −60 and 60 mV membrane voltages, respectively, at different Ca concentrations (). The small insignificant difference between the calculated ratio at 1.5 mM Ca and low divalent is consistent with the high affinity of Ca to the TRPL channel (). Importantly, the calculated ratio at divalent-free conditions is about twofold larger than under low divalent conditions, reflecting the almost linearity of the I-V curve (, red). Together, the study of the native TRPL channels indicates that their voltage dependence is not an intrinsic property but arises from Ca block ().
In the present study, the mechanism underlying the voltage dependence of the TRPL channel was investigated for the first time. The detailed analyses of single channel and whole cell currents indicated that the voltage dependence of the TRPL channel is not an intrinsic property but arises from open channel block by Ca. The ability to robustly activate these channels at physiologically relevant negative membrane potentials by reducing Ca levels was manifested by the appearance of a large inward current in the expressed and native TRPL channels. Previous reports showed only outward rectification of the TRPL channels (). However, and show that both the expressed and native TRPL channels revealed a pronounced conductance also at negative membrane potential in Ca free conditions, which can be blocked by the presence of Ca on both sides of the plasma membrane. This observation fits well with notion of voltage-dependent block by a permeable charged blocking agent that permeates the channel from both outside and inside the channel, while only depolarization can relieve this block (; see also ). Enhanced channel openings at positive membrane potential of TRPL in the presence of divalent cations is fully consistent with the notion that Ca open channel block underlies the outward rectification of these channels, because depolarization removes Ca channel block. Similarly, channel openings at negative membrane potential in divalent-free conditions, when this block is reduced, support the above notion. The voltage-dependent Ca block of the native TRPL channel () is reminiscent of a very similar property of cyclic nucleotide-gated (CNG) channels, which was studied in great detail by Kaupp and colleagues (). This voltage-dependent blockage of the open channel is characteristic of positively charged, permeable channel blocker, where blockage efficacy increases when the membrane voltage is made less positive because access of Ca to the blocking site in the channel lumen is facilitated (). The close similarity of the Ca open channel block phenomenon between the TRPL channel and the thoroughly investigated CNG channels strongly supports the notion that the Ca blocks the native TRPL channel by an open channel block mechanism.
Although the TRPL channel is a calmodulin (CaM) binding protein (; ), the blocking effect of Ca on the channel most likely did not involve CaM because the blocking effect of Ca was also found after application of Mg and Ba. This conclusion is supported by the findings of Flockerzi and colleagues who found that TRPL channel block by Ca was not affected by calmidazolium, a known blocker of CaM and suggested that Ca blocks the TRPL channel directly ().
The difference in the mechanism underlying the voltage dependence of TRPC channels (i.e., TRPL) and the intrinsic voltage dependence of TRPV and TRPM channels (i.e., TRPV1 and TRPM8; ) may arise from the difference in the amino acid sequence at the pore region between the TRPC and TRPV&TRPM channels, which allows open channel block by Ca in TRPL channels but not in TRPV channels (). Taken together, the gating mechanism of the TRPL channel does not involve activation by voltage changes as suggested for TRPV and TRPM channels.
The photoreceptor cell has two functional light-activated channels, TRP and TRPL. The TRP channels have a small (∼4 pS) conductance (), while the TRPL channels have much larger conductance of 110 pS (). The TRPL channels contribute ∼40% to the total light-induced current in dark-adapted flies in response to dim or medium intensity lights (). At medium light intensities, the LIC is a sum of the current flow through both the TRP and TRPL channels (; ). In contrast, during intense light, the LIC of WT flies is very similar to that of the mutant (; ), indicating that the TRPL channels are inactive. This observation is readily explained by the higher Ca affinity of the TRPL channels relative to the TRP channels (; ). (The lower Ca affinity of the TRP relative to the TRPL channel is clearly reflected in I-V curves measured at nominal external Ca [∼10 μM external Ca] in the relevant mutants. While the I-V curve of the mutant shows strong outward rectification and minimal inward current at nominal external Ca, the mutant [or WT flies] shows a strong inward and outward rectification at the same external Ca concentration [; ].) Since the TRPL channel has large single channel conductance, it greatly contributes to the noise of the LIC (). It is therefore expected, according to an open channel block mechanism, that at intense light, when all TRPL channels are blocked, the noise level will be reduced. In contrast, at medium light intensities, when a large fraction of the TRPL channels is still available for light activation, the noise level will be higher. Indeed, the study of Hardie and colleagues () shows that the noise level of the LIC induced by medium intensity light is significantly larger in WT than in the mutant (with no TRPL channels). Taken together, the signal to noise ratio of the light response is improved during intense light by Ca open channel block of the TRPL channel.
It is interesting to note that divalent open channel block that largely reduces the single channel conductance is a property of CNG channels of vertebrate photoreceptors, indicating that divalent open channel block is a general strategy of increasing the signal to noise ratio of photoreceptor cells in the animal kingdom (). |