Patent Application
20150252417
The present invention relates, in general, to gene expression profiles, and in particular, to a gene expression profile (e.g., a peripheral blood gene expression profile) of an environmental exposure, ionizing radiation. The invention further relates to methods of screening patients for radiation exposure based on gene expression profiling and to kits suitable for use in such methods.
Invasive procedures are often required for accurate screening and diagnosis of common medical conditions (Boolchand et al, Ann. Intern. Med. 145:654-659 (2006)). Examination of the peripheral blood often suffices to establish certain diagnoses, such as chronic lymphocytic leukemia (Damle et al, Blood Epub Ahead of Print (2007)), which afflicts the circulating lymphocyte directly. Measurement of total white blood cell counts and the WBC differential (e.g. neutrophils, lymphocytes, monocytes) is routinely performed in medical practice and can facilitate many diagnoses (e.g. bacterial or viral infection). Recently, it has been suggested that gene expression profiling of peripheral blood cells, particularly lymphocytes, can serve as sensitive tool to assess for the presence of certain diseases, such as systemic lupus erythematosus, rheumatoid arthritis, neurologic disease, viral and bacterial infections, breast cancer, atherosclerosis and environmental exposures, including tobacco smoke (Mandel et al, Lupus 15:451-456 (2006), Heller et al, Proc. Natl. Acad. Sci. USA 94:2150-2155 (1997), Edwards et al, Mol. Med. 13:40-58 (2007), Baird, Stroke 38:694-698 (2007), Rubins et al, Proc. Natl. Acad. Sci. USA 101:15190-15195 (2004), Martin et al, Proc. Natl. Acad. Sci. USA 98:2646-2651 (2001), Patino et al, Proc. Natl. Acad. Sci. USA 102:3423-3428 (2005), Lampe et al, Cancer Epidemiol. Biomarkers Prev. 13:445-453 (2004), Ramilo et al, Blood 109:2066-2077 (2007)). Results from these studies suggest that patterns of gene expression within circulating PB cells can distinguish individuals afflicted by these conditions from those who are not (Mandel et al, Lupus 15:451-456 (2006), Heller et al, Proc. Natl. Acad. Sci. USA 94:2150-2155 (1997), Edwards et al, Mol. Med. 13:40-58 (2007), Baird, Stroke 38:694-698 (2007), Rubins et al, Proc. Natl. Acad. Sci. USA 101:15190-15195 (2004), Martin et al, Proc. Natl. Acad. Sci. USA 98:2646-2651 (2001), Patino et al, Proc. Natl. Acad. Sci. USA 102:3423-3428 (2005), Lampe et al, Cancer Epidemiol. Biomarkers Prey. 13:445-453 (2004), Ramilo et al, Blood 109:2066-2077 (2007)). It has, therefore, been suggested that PB gene expression profiling has potential utility in the screening for diseases and environmental exposures.
Any consideration of applying PB gene expression profiles for the detection of disease or environmental exposures requires a determination of the impact of PB cellular composition, time, gender, and genotype on PB gene expression (Lampe et al, Cancer Epidemiol. Biomarkers Prev. 13:445-453 (2004), Ramilo et al, Blood 109:2066-2077 (2007), Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003), Yan et al, Science 297:1143 (2002)). Additionally, it is unclear whether PB gene expression profiles that have been associated with various medical conditions are specific for that phenotype, or rather reflect a generalized response to genotoxic stress. Examination of the specificity of PB gene expression profiles in response to different stimuli and the durability of these signatures over time is critical to the translation of this strategy into clinical practice.
Ionizing radiation represents a particularly important environmental hazard, which, at lowest dose exposures, causes little acute health effects (Kaiser, Science 302:378 (2003)) and, at higher dose exposures, can cause acute radiation syndrome and death (Wasalenko et al, Ann. Int. Med. 140:1037-1051 (2004), Mettler et al, N. Engl. J. Med. 346:1554-1561 (2002), Dainiak, Exp. Hematol. 30:513-528 (2002)). Numerous studies have been performed to attempt to understand the biologic effects of ionizing radiation in humans. Specific mutations in p53 and HPRT have been identified in somatic cells from survivors of the Hiroshima and Nagasaki atomic bombings (Iwamoto et al, J. Natl. Canc. Inst. 90:1167-1168 (1998), Hirai et al, Mutant Res. 329:183-196 (1995), Takeshima et al, Lancet 342:1520-1521 (1993), Neel et al, Annu. Rev. Genet. 24:327-362 (1990)).
Gene expression analyses have been performed on human tumor cells, cell lines, and peripheral blood from small numbers of irradiated patients in order to identify specific genes that are involved in the response to radiation injury (Jen et al, Genome Res. 13:2092-2100 (2003), Amundson et al, Radiat. Res. 154:342-346 (2000), Amundson et al, Radiat. Res. 156:657-661 (2001), Falt et al, Carcinogenesis 24:1837-1845 (2003), Amundson et al, Cancer Res. 64:6368-6371 (2004)). Recently, public health focus has centered on the development of capabilities to accurately screen large numbers of people for radiation exposure in light of the anticipated use of radiological or nuclear materials by terrorists to produce “dirty bombs” or “improvised nuclear devices” (Wasalenko et al, Ann. Int. Med. 140:1037-1051 (2004), Mettler et al, N. Engl. J. Med. 346:1554-1561 (2002), Dainiak, Exp. Hematol. 30:513-528 (2002)).
A method of screening humans for environmental exposure has been suggested. This method relies on the identification of patterns of gene expression, or metagenes in PB cells that accurately distinguish between irradiated and non-irradiated individuals (Dressman et al, PLoS Med. 4:690-701 (2007)). Metagenes can be identified in the PB that distinguish different levels of exposure with an accuracy of 96% (Dressman et al, PLoS Med. 4:690-701 (2007)).
The present invention results, at least in part, from studies designed to evaluate the specificity of PB gene expression signatures and to determine the influence of genetic variation and time on the performance of the signature. The invention also results from studies in which an examination has been made of the possibility of “training” a biodosimeter in three model systems simultaneously under the hypothesis that a biodosimeter that is functional in all three systems has a higher likelihood of performing well in the population of interest. The results of these studies indicate that this approach represents a viable strategy for identifying environmental exposures and one that can be employed for screening populations of affected individuals.
The present invention relates generally to gene expression profiles. More specifically, the invention relates to a gene expression profile of an environmental exposure, ionizing radiation. The invention further relates to a method of screening patients for radiation exposure based on gene expression profiling and to kits suitable for use in such methods.
Objects and advantages of the present invention will be clear from the description that follows.
FIG. 8-8.—Concurrent gene behavior. Each of the 9150 genes with mouse-human analogs were tested for correlation with radiation exposure dose. The x-axis shows that computed correlation in the human TBI patients and the y-axis shows that correlation in human ex vivo. Red lines indicate significant p-values after Bonferroni correction for multiple testing. The color spots indicate correlation in mice, with red indicating positive correlation and green negative. The size and brightness of the spot indicates the level of correlation for that gene in mice. The generally greener bottom left corner and redder upper right corner indicate a general agreement between mice and human data, but the presence of green spots in the upper right indicates that individual genes may behave significantly differently in different model systems.
The present invention results, at least in part, from the demonstration that exposure to ionizing radiation induces a pronounced and characteristic alteration in PB gene expression. The expression profiles disclosed herein provide basis for a method of screening a heterogeneous human population, for example, in the event of a radiological or nuclear event.
Examples of gene expression profiles that can be used distinguish radiation status in humans include those set forth in Tables 7 and 9 (and
A preferred profile is set forth in Table 11 (see response genes FDXR, ASPA, RFC4, METTL8, RASL12, ASTN2, RASA4, TRIB2, BBC3, RPA1, Gna15, H2AFV, CEBPB, CDKN1A, PRIM1, NINJ1, BAX, HIST1H3D, HIST1H2BH, DDB2, BCL11B, FAM134C and LAPTM5—details of the response genes included in Table 11 are provided in Table 7 and/or 9, with the exception of FDXR and HIST1H2BH, details for which are provided in Table 11). Subsets of the signature set forth in Table 11 (e.g., comprising at least 5 or at least 10 or at least 20 response genes) are potentially suitable for use in accordance with the present invention.
In one embodiment, the invention relates to a method screening a patient for radiation exposure by collecting a sample (e.g., PB) from the patient and isolating mononuclear cells therefrom. RNA can be extracted from the mononuclear cells using standard techniques, including those described in the Examples below. The extracted RNA can be amplified and suitable probes prepared (see Examples and Dressman et al, PLoS Med. 4:690-701 (2007)). Gene expression levels can then be determined using, for example, microarray techniques (see Examples and Dressman et al, PLoS Med. 4:690-701 (2007)).
In accordance with one embodiment of the invention, a patient that displays the gene expression profile set forth in Table 7 is a patient that has been exposed to radiation (e.g., about 6 hours prior to PB collection). While the 25 genes set forth in Table 7 constitute one signature suitable for use is distinguishing radiation status, the invention also includes methods based on the use of signatures comprising the following: H200000088, H200008365, H200011577, H200014719, H200016323, H300000421, H300003103, H300010830, H300015667, H300019371, H300020858, H300021118, H300022025. Other subsets of the signature set forth in Table 7 (e.g., comprising at least 5 or at least 10 genes) are potentially suitable for use in accordance with the present invention.
In accordance with a preferred embodiment of the invention, a patient that displays the gene expression profile set forth in Table 11 is a patient that has been exposed to radiation (e.g., about 6 hours prior to PB collection). While the 23 response genes set forth Table 11 constitute one signature suitable for use in distinguishing radiation status the invention also includes methods based on the use of signatures comprising subsets of the response gene signature set forth in Table 11 (e.g., subsets comprising at least 5, 10 or 20 response genes).
The development of a biodosimeter for the purpose of triaging patients after a major accident or attack must necessarily be conducted without samples from otherwise healthy people who have been exposed. As described herein, model systems can be used to determine the behavior of putative biomarkers in a human population. A biodosimeter that is functional in multiple systems can be expected to perform well in the population of interest.
The studies described in Example 2 indicate that there is some gene-level concordance in the response to radiation among model systems: mouse, human ex vivo, and human TBI. However, that concordance is generally mild, and the generation of a biodosimeter based on just one model system for use in the other two leads to poor performance. To address this issue, a variable-selection regression model has been used that includes training data from all samples (see Example 2). This approach has resulted in a predictive model that differentiates dose in all of the model systems.
It is expected that this predictive model will perform adequately in stratifying subjects by dose in the event of an accident or attack leading to radiation exposure in an otherwise healthy human population. However, as evidenced by the existence of significant predictive genes in all three of the model systems that are not relevant in the other systems (see Example 2), it is expected that there are genes that can be used to advantage in a biodosimeter that cannot be identified with the model-systems approach. Accordingly, it is likely—given real data from such an event—that by inclusion of new genes, the accuracy of the biodosimeter can be improved upon in an exposed, otherwise healthy human population.
Finally, while the biodosimeter described performs well on microarray data, a high throughput, low cost gene expression platform may be preferred for use in the field. It is understood that at least some of the genes in the predictor may not translate well between platforms. In order to alleviate this potential problem, a relatively large set of predictors have been retained for translation. The genes set forth, for example, in Table 9 (see also
While the expression profiles described herein are highly predictive of radiation status, sex differences can contribute to characteristically distinct molecular responses to radiation, for example at low exposure levels (e.g., about 50 cGy). Accordingly, use of gender specific assays can be advantageous, for example, at low levels of exposure.
As shown in Example 1 that follows, the time of PB collection following radiation exposure does not significantly impact the accuracy of PB signatures to predict radiation status or distinguish different levels of exposure. While time as a single variable does not lessen the accuracy in distinguishing irradiated from non-irradiated individuals, the content of the genes which comprise the PB signature can change as a function of time. Thus, while PB predictors of radiation exposure can change over time, PB signatures can continuously be identified (e.g., through 7 days) that are highly accurate at predicting radiation status and distinguishing different levels of exposure.
The invention also relates to reagents and kits suitable for use in practicing the above-described methods. Kit components can vary, however, examples of components include an array probe of nucleic acids in which the genes listed in Table 7 and/or Table 9 (see also
Certain aspects of the invention can be described in greater detail in the non-limiting Examples that follow. (See also Dressman et al, PLoS Med. 4:690-701 (2007) and U.S. application Ser. No. 12/736,393)).
Ten to 11 week old male and female C57BI6 and female BALB/c mice (Jackson Laboratory, Bar Harbor, Me.) were housed at the Duke Cancer Center Isolation Facility under regulations approved by the Duke University Animal Care and Use Committee. Between 5-10 mice/group were given total body irradiation (TBI) with a Cs137 source at an average of 660 cGy/min at doses of 50, 200, or 1000 cGy as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). Six hours, 24 hours, or 7 days post-TBI, approximately 500 μl peripheral blood was collected by retro-orbital bleed from both irradiated and control mice. PB mononuclear cells (PB MNCs) were isolated for total RNA extractions. Total RNA was extracted with Qiagen RNAeasy Mini Kits as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). RNA quality was assayed using an Agilent Bioanalyzer 2100 (Agilent Technologies, Inc., Palo Alto, Calif.).
Ten C57BI6 female mice were given intraperitoneal injections of a 100 μg of lipopolysaccharide endotoxin (LPS) from E. coli 055:B5 (Sigma-Aldrich, St. Louis, Mo.) to induce sepsis syndrome as previously described (Hick et al, J. Immunol. 177:169-176 (2006)). Peripheral blood was collected 6 h later from treated and control mice, and RNA was processed as described in the irradiation studies.
With approval from the Duke University Institutional Review Board (IRB), between 5-12 mL of peripheral blood was collected from patients prior to and 6 hrs following total body irradiation with 150 to 200 cGy as part of their pre-transplantation conditioning (Dressman et al, PLoS Med. 4:690-701 (2007)). For additional comparison, peripheral blood was obtained from healthy volunteers and an additional cohort of patients prior to and 6 hrs following the initiation of alkylator-based chemotherapy alone (without radiotherapy). All patients and healthy volunteers who participated in this study provided written informed consent prior to enrollment, as per the Duke IRB guidelines. PB MNCs and total RNA were isolated from the blood using the identical methods as described for collection of murine cells and RNA.
Mouse and human oligonucleotide arrays were printed at the Duke Microarray Facility using Operon's Mouse Genome Oligo sets (version 3.0 and version 4.0) and Operon's Human Genome Oligo set (version 3.0 and version 4.0). Data generated from Operon's Mouse and Human version 3 was previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). Operon's Mouse Genome Oligo set (version 4.0) (https://www.operon.com/arrays/oliqosets mouse.php) contains 35,852 oligonucleotide probes representing 25,000 genes and approximately 38,000 transcripts. Operon's Human Genome Oligo set (version 4.0) (https://www.operon.com/arrays/oliqosets human.php) contains 35,035 oligonucleotide probes, representing approximately 25,100 unique genes and 39,600 transcripts. In order to compare across versions of the Operon oligo sets, Operon provided a map that matched the probes from both versions and only those oligonucleotides that overlapped between versions 3.0 and 4.0 were used in the analysis.
Briefly, MNCs were pelleted, and total RNA was isolated using the RNAeasy mini spin column (Dressman et al, PLoS Med. 4:690-701 (2007)). Total RNA from each sample (mouse or human) and the universal reference RNA (Universal Human or Mouse Reference RNA, Stratagene, http://www.strataqene.com) were amplified and used in probe preparation as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). The sample (mouse or human) was labeled with Cy5 and the reference (mouse or human) was labeled with Cy3. The reference RNA allows for the signal for each gene to be normalized to its own unique factor allowing comparisons of gene expression across multiple samples. This serves as a normalization control for two-color microarrays and an internal standardization for the arrays. Amplification, probe preparation and hybridization protocols were performed as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)) and each condition examined had multiple replicates analyzed (n=3-18 per mouse condition and n=18-36 per human condition). Detailed protocols are available on the Duke Microarray Facility web site (http://microarray.genome.duke.edu/services/spotted-arrays/protocols).
Genespring GX 7.3 (Agilent Technologies) was used to perform initial data filtering in which spots whose signal intensities below 70 in either the Cy3 or Cy5 channel were removed. For each analysis, only those samples in that analysis were used in the filtering process. To compare data from previously published results (Dressman et al, PLoS Med. 4:690-701 (2007)), only those probes were used that mapped to each other across the version 3.0 and version 4.0 arrays. To then account for missing values, PAM software (http://www-stat.stanford.edu/˜tibs/PAM/) was used to impute missing values. k-nearest neighbor was used where missing values were imputed using a k-nearest neighbor average in gene space. In the analysis approach in which all samples were included, lowess normalization of the data followed by batch effect removal using 2-way mixed model ANOVA (Partek Incorporated) was performed. Gene expression profiles of dose response were used in a supervised analysis using binary regression methodologies as described previously (Dressman et al, PLoS Med. 4:690-701 (2007)). Prediction analysis of the expression data was performed using MATLAB software as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). When predicting levels of radiation exposure, gene selection and identification is based on training the data and finding those genes most highly correlated to response. Each signature summarizes its constituent genes as a single expression profile and is here derived as the first principal component of that set of genes (the factor corresponding to the largest singular value), as determined by a singular value decomposition. Given a training set of expression vectors (of values across metagenes) representing two biological states, a binary probit regression model is estimated using Bayesian methods. Bayesian fitting of binary probit regression models to the training data then permits an assessment of the relevance of the metagene signatures in within-sample classification, and estimation and uncertainty assessments for the binary regression weights mapping metagenes to probabilities of radiation exposure. To internally validate the predictive capacity of the metagene profiles, leave-one-out cross validation studies were performed as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). A leave one out cross validation involves removing one sample from the dataset, using the remaining samples to develop the model, and then predicting the status of the held out sample. This is then repeated for each sample in the dataset. This approach was utilized as previously described (Dressman et al, PLoS Med. 4:690-701 (2007)). A ROC curve analysis was used to define a cut-off for sensitivity and specificity in the predictive models of radiation. Genes found to be predictive of radiation dose were characterized utilizing an in-house program, GATHER (http://meddb01.duhs.duke.edu/qather/). GATHER quantifies the evidence supporting the association between a gene group and an annotation using a Bayes factor (Pournara et al, BMC Bioinformatics 23:1-20 (2007)). All microarray data files can be found at http://data.cgt.duke.edu/ChuteRadiation.php and at gene expression omnibus website (GEO [http://www.ncbi.nlm.nih.gov/geo], accession number GSE10640).
In a previous study, it was demonstrated that PB collected from a single strain and gender of mice, at a single time point, contained patterns of gene expression that predicted both prior radiation exposure and distinguished different levels of radiation exposure with a high degree of accuracy (Dressman et al, PLoS Med. 4:690-701 (2007)). In this study, a determination was made as to whether PB gene expression signatures could be identified that predict radiation exposure status within a population that was heterogeneous for genotype, gender and time of sampling. It was found that a clear pattern of gene expression could be identified within this heterogeneous population of mice that distinguished non-irradiated animals from those irradiated with 50 cGy, 200 cGy, and 1000 cGy (
A determination was then made as to the extent to which variables within a heterogeneous population can limit the accuracy of PB gene expression profiling. In order to address the impact of sex difference, healthy adult male and female C57BI6 mice were irradiated with 50 cGy, 200 cGy, and 1000 cGy and PB was collected at 6 hours post-irradiation, along with PB from non-irradiated control mice (n=7-10 per group). Patterns of gene expression could be identified in the PB of both male and female mice that appeared to distinguish radiation exposure status (
Since the human population is genetically diverse, an examination was next made to determine whether gene expression signatures of radiation exposure could accurately predict the status of mice across different genotypes. PB was collected from C57BI6 and BALB/c mice at 6 hours following 50 cGy, 200 cGy or 1000 cGy. It was possible to identify patterns of gene expression which appeared to distinguish the different levels of radiation from the non-irradiated controls within each strain (
PB responses to environmental exposures may change over time as a function of changes in PB cellular composition and cellular responses themselves. Patterns of gene expression were identified in the PB of C57BI6 female mice at 6 hrs, 24 hrs and 7 days post-irradiation which appeared to distinguish the 3 different levels of radiation versus non-irradiated mice (
In addition to inter-individual variations (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)), human populations are heterogeneous with respect to health status and medical conditions. Therefore, it is critical to determine whether PB gene expression profiles of radiation response are specific to radiation exposure itself or whether these signatures are potentially confounded by other genotoxic stresses. The choice was made to compare the PB gene expression response to ionizing radiation exposure with that of gram-negative bacterial sepsis, since this syndrome can be expected to induce similar multiorgan toxicity as is observed following radiation injury (Wasalenko et al, Ann. Int. Med. 140:1037-1051 (2004), Mettler et al, N. Engl. J. Med. 346:1554-1561 (2002), Dainiak, Exp. Hematol. 30:513-528 (2002), Inoue et al, FASEB J. 20:533-535 (2006)). A pattern of gene expression could be identified which effectively distinguished female C57BI6 mice treated with Escherichia coli-derived lipopolysaccharide (LPS), experiencing sepsis syndrome, from untreated female C57BI6 mice (
BLE 5
tp://omad.operon.com)
peron Oligo ID
200003295
300009870
300005418
200001838
200000053
200004157
300005305
300006479
200004765
300005673
300005674
300001891
300005166
200005576
300020771
300011591
200007439
300012693
300012210
200014281
300009340
200004564
300000152
200006264
300000077
indicates data missing or illegible when filed
In order to extend the analysis of PB signature specificity to humans, PB was collected from a population of healthy individuals (n=18), patients who had undergone total body irradiation as conditioning prior to hematopoietic stem cell transplantation (n=47) and patients who had undergone alkylator-based chemotherapy conditioning alone (n=41). RNA of sufficient quality was available from 18 healthy donor samples, 36 pre-irradiated patients, 34 post-irradiated patients, 36 pre-chemotherapy treatment patients and 32 post-chemotherapy patients (Table 6). A supervised binary regression analysis identified a metagene profile of 25 genes that distinguished the healthy individuals and the non-irradiated patients from the irradiated patients (
In order to test the specificity of this PB signature of human radiation response, its accuracy was next tested in predicting the status of patients who had undergone chemotherapy treatment alone. This signature correctly predicted 89% of the non-irradiated, pre-chemotherapy patients as non-irradiated and 75% of the chemotherapy-treated patients as non-irradiated (
BLE 8
ttp://omad.operon.com)
eron
go_ID
00001454
00002954
00004993
00002479
00020334
00002535
00002231
00002230
00001588
00001821
00001397
00008401
00022877
00018428
00022441
00014949
0006902
00001600
00007970
00019724
00004653
00012785
00010704
00017278
00005078
indicates data missing or illegible when filed
Summarizing, numerous studies now highlight the power of gene expression profiling to characterize the biological phenotype of complex diseases. The potential clinical utility of gene expression profiles has been shown in cancer research, in which the identification of patterns of gene expression within tumors has led to the characterization of tumor subtypes, prognostic categories and prediction of therapeutic response (Potti et al, N. Engl. J. Med. 355:570-580 (2006), Cheng et al, J. Clin. Oncol. 24:4594-4602 (2006), Potti et al, Nat. Med. 12:1294-1300 (2006), Nevins et al, Nat. Rev. Genet. 8:601-609 (2007), Alizadeh et al, Nature 403:503-511 (2000)). Beyond analysis of tumor tissues, it has also been suggested that gene expression profiling of the peripheral blood can provide indication of infections, cancer, heart disease, allograft rejection, environmental exposures and as a means of biological threat detection (Mandel et al, Lupus 15:451-456 (2006), Heller et al, Proc. Natl. Acad. Sci. USA 94:2150-2155 (1997), Edwards et al, Mol. Med. 13:40-58 (2007), Baird, Stroke 38:694-698 (2007), Rubins et al, Proc. Natl. Acad. Sci. USA 101:15190-15195 (2004), Martin et al, Proc. Natl. Acad. Sci. USA 98:2646-2651 (2001), Patino et al, Proc. Natl. Acad. Sci. USA 102:3423-3428 (2005), Lampe et al, Cancer Epidemiol. Biomarkers Prey. 13:445-453 (2004), Ramilo et al, Blood 109:2066-2077 (2007), Horwitz et al, Circulation 110:3815-3821 (2004), Lin et al, Clinic Chem. 49:1045-1049 (2003)). While the concept of PB cells as sentinels of disease is not new, it remains unclear whether PB gene expression profiles that have been associated with various conditions are specific for those diseases or rather reflect a common molecular response to a variety of genotoxic stresses. Given the dynamic nature of the cellular composition of PB blood (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)) and the complexity of cellular responses over time (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)), the durability of PB signatures over time is also uncertain and could affect the diagnostic utility of this approach for public health screening.
A purpose of the studies described above was to address the capacity for PB gene expression profiles to distinguish an environmental exposure, in this case ionizing radiation, versus other medical conditions and to examine the impact of time, gender and genotype on the accuracy of these profiles. It was found that PB gene expression signatures can be identified which accurately predict irradiated from non-irradiated mice and distinguish different levels of radiation exposure, all within a heterogeneous population with respect to gender, genotype and time from exposure. These results suggest the potential for PB gene expression profiling to be applied successfully in the screening for an environmental exposure. Previous studies have indicated that inter-individual variation in gene expression occurs within healthy individuals (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)) and may therefore limit the accuracy of PB gene expression profiling to detect diseases or exposures. The results provided here demonstrate that the environmental exposure tested here, ionizing radiation, induced a pronounced and characteristic alteration in PB gene expression such that a PB expression profile was highly predictive of radiation status in a population with variable gender, genotype and time of analysis. From a practical standpoint, these data suggest the potential utility of this approach for biodosimetric screening of a heterogeneous human population in the event of a purposeful or accidental radiological or nuclear event (Wasalenko et al, Ann. Int. Med. 140:1037-1051 (2004), Mettler et al, N. Engl. J. Med. 346:1554-1561 (2002), Dainiak, Exp. Hematol. 30:513-528 (2002)).
This study revealed that sex differences can impact the accuracy of this approach, particularly in distinguishing mice exposed to lower dose irradiation from non-irradiated controls. These results imply that aspects of the PB response to ionizing radiation are specified by sex-associated genes. Whitney et al (Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)) previously showed that sex differences were associated with variation in PB autosomal gene expression in healthy individuals. The instant study suggests that sex differences may contribute to characteristically distinct PB molecular responses to environmental stress (radiation) and the accuracy of PB gene expression profiling for medical screening can be affected by sex. These sex-related differences in PB response to ionizing radiation are perhaps illustrated by the fact that only 2 genes overlapped between the male and female PB signatures of 50 cGy (Ccng1 and Dda3).
Interestingly, differences in genotype did not significantly impact the accuracy of the PB gene expression signatures to distinguish radiation response such that PB signatures from C57BI6 mice displayed 100% accuracy in predicting the status of BALB/c mice and vice versa. This observation demonstrates that, while genotype differences can account for some variation in PB gene expression (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)), the alterations in PB gene expression induced by 3 different levels of radiation exposure are such that PB expression profiling is highly accurate in distinguishing all irradiated mice across different genotypes. Very few genes were found in common between the 2 strains of mice at each level of radiation exposure, indicating that diverse sets of genes contribute to the PB response to radiation and that unique sets of genes can be identified which are predictive of radiation response.
The time of PB collection following radiation exposure had no significant impact on the accuracy of PB signatures to predict radiation status or distinguish different levels of exposure. First, the accuracy of PB signatures to predict radiation status and distinguish different levels of radiation exposure did not decay over time. Second, when we applied a PB signature from a single time point (6 hrs) against PB samples collected from mice at other time points (24 hr and 7 days), the accuracy of the prediction remained 100% in all cases. Therefore, time as a single variable did not lessen the accuracy of this approach to distinguish irradiated from non-irradiated animals. However, the content of the genes which comprised the PB signatures changed significantly as a function of time and <20% of the genes overlapped between the PB signatures of radiation at 6 hr, 24 hr, and 7 days. Taken together, these data indicate that PB predictors of radiation response do change over time, but PB signatures can continuously be identified through 7 days that are highly accurate at predicting radiation status and distinguishing different levels of radiation exposure. From a practical perspective, these results suggest that the application of a single reference set of “radiation response” genes would be unlikely to provide the most sensitive screen for radiation exposure over time. Conversely, reference lists of PB genes that are specific for different time points could be applied in the screening for radiation exposure provided that the time of exposure was known.
A critical question to be addressed in the development of PB gene expression profiling to detect medical conditions or exposures is the specificity of PB gene expression changes in response to genotoxic stresses. The PB signatures of 3 different doses of radiation displayed 100% accuracy in identifying septic animals as non-irradiated and the PB signature of sepsis was also 100% accurate in identifying irradiated mice as non-septic. These results demonstrate specificity in the PB responses to ionizing radiation and sepsis. These data also provide in vivo validation of a prior report by Boldrick et al (Proc. Natl. Acad. Sci. USA 99:972-977 (2002)) in which human PB mononuclear cells were found to have a stereotypic response to LPS exposure in vitro and specific alterations in gene expression were observed in response to different strains of bacteria (Boldrick et al, Proc. Natl. Acad. Sci. USA 99:972-977 (2002)). Ramilo et al. also recently reported that distinct patterns of PB gene expression can be identified among patients with different bacterial infections (Ramilo et al, Blood 109:2066-2077 (2007)). No genes were found to be in common between the PB signatures of radiation exposure and the PB signature of gram negative sepsis. Taken together, the results demonstrate that the in vivo PB molecular responses to ionizing radiation and bacterial sepsis are quite distinct and can be utilized to distinguish one condition from the other with a high level of accuracy.
The analyses of expression signatures in human patients demonstrated that it is possible to utilize PB gene expression profiles to distinguish individuals who have been exposed to an environmental hazard, ionizing radiation, within a heterogeneous human population with a high level of accuracy. It will be important to further test the accuracy of this PB predictor of human radiation exposure in a human population exposed to lower dose irradiation (e.g. 0.1-1 cGy), as might be expected via occupational exposures (e.g. radiology technicians, nuclear power plant workers) (Seierstad et al, Radiat. Prot. Dosimetry 123:246-249 (2007), Moore et al, Radiat. Res. 148:463-475 (1997), Einstein et al, Circulation 116:1290-1305 (2007)). A potential pitfall in the clinical application of PB gene expression profiling would be that variations in PB gene expression in people would be such that it might be difficult to distinguish the effects of a given exposure or medical condition from expected background alterations in gene expression (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)). However, Whitney et al (Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)) showed that the alterations in PB gene expression observed in patients with lymphoma or bacterial infection was significantly greater than the relatively narrow variation observed in healthy individuals (Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003)). This study confirms that PB gene expression profiles can be successfully applied to detect a specific exposure in a heterogeneous human population and that inter-individual differences in PB gene expression do not significantly confound the utility of this approach.
It was also shown that unique PB gene expression profiles can be identified which distinguish chemotherapy-treated patients versus patients who had not received chemotherapy with an overall accuracy of 81% and 78%, respectively. Similar to the PB signature of radiation, the PB signature of chemotherapy demonstrated accuracy and specificity in distinguishing healthy individuals and pre-irradiated patients (100% and 92% accuracy, respectively). However, the accuracy of the PB signature of chemotherapy was more limited when tested against patients who received radiation conditioning (62%). This observation provides the basis for further investigation as to which families of genes may be represented in both the PB molecular response to radiation and chemotherapy. However, since all 12 of the post-irradiation patients whose status was mispredicted by the PB chemotherapy signature had received combination chemotherapy within the prior year, the true specificity of this PB signature of chemotherapy cannot be addressed via this comparison. Additional patients are currently being enrolled to this study who have not undergone prior chemotherapy to further test the specificity of a PB metagene of chemotherapy treatment.
Peripheral blood is a readily accessible source of tissue which has the potential to provide a window to the presence of disease or exposures. Early studies applying PB gene expression analysis have demonstrated that this approach is sensitive for the detection of patterns of gene expression in association with a variety of medical conditions (Mandel et al, Lupus 15:451-456 (2006), Heller et al, Proc. Natl. Acad. Sci. USA 94:2150-2155 (1997), Edwards et al, Mol. Med. 13:40-58 (2007), Baird, Stroke 38:694-698 (2007), Rubins et al, Proc. Natl. Acad. Sci. USA 101:15190-15195 (2004), Martin et al, Proc. Natl. Acad. Sci. USA 98:2646-2651 (2001), Patino et al, Proc. Natl. Acad. Sci. USA 102:3423-3428 (2005), Lampe et al, Cancer Epidemiol. Biomarkers Prev. 13:445-453 (2004), Ramilo et al, Blood 109:2066-2077 (2007), Whitney et al, Proc. Natl. Acad. Sci. USA 101:1896-1901 (2003), Dressman et al, PLoS Med. 4:690-701 (2007)). It remains to be determined whether PB gene expression profiles can be successfully applied in medical practice or public health screening for the early detection of specific diseases or environmental exposures. The present results demonstrate that PB gene expression profiles can be identified in mice and humans which are specific, accurate over time, and not confounded by inter-individual differences.
Gene expression in peripheral blood was measured with the Affymetrix mouse 430A 2.0 microarray and Affymetrix human U133A 2.0 microarray. Because there is interest in creating predictors that are consistent for all model systems, the gene list was filtered to include only those genes with mouse-human analogs. For the results presented, analogs were mapped using Chip Comparer (Yao G. Chip comparer. 2005. http://chipcomparer.genome.duke.edu (accessed Oct. 3, 2011)), though results were similar using the approach of matching gene names from the Affymetrix annotation files. Annotation mapping resulted in 9150 genes with matching analogs in both mouse and human microarrays.
In order to assess the level of concordant information among mouse, human ex vivo, and human TBI, correlations between each gene and known radiation exposure level (nonparametric Kendall correlation) were tested. This procedure resulted in a set of correlations and p-values for each gene in each of the three model systems. If genes are behaving consistently in response to radiation, then general agreement in these correlations is expected. There are six time points for the mouse study and two for the human ex vivo study. Because it is not known how closely aligned the temporal responses of mice and humans are, the level of agreement between these correlation values was examined for all possible pairings of times points. In addition, all data was tested without regard to time. The highest level of agreement is from comparing human TBI to 24-h human ex vivo. The highest level of agreement between mouse and human TBI is at 6 h in the mice. In general, there is much higher agreement between human TBI and human ex vivo data sets than there is between mouse and either human data set. It is difficult to determine whether this is caused by fundamental differences in the responses of mice and humans to radiation exposure or caused by difficulties in mapping mouse-human orthologs.
Model building was restricted to the 169 genes with absolute gene-dose correlation greater than 0.2 in all three of the model systems. Because there is interest in a limited list of genes for an eventual diagnostic, use was made of a variable-selection prior distribution on the linear regression coefficients to limit the number of genes in the model. There are many resultant models that are consistent with the data. Therefore, model averaging was used to control for uncertainty in the choice of inclusion variables.
A single biosignature has been built that stratifies radiation-exposed samples from three model systems by dose. The model systems used were mouse C57BI6, human ex vivo, and hospitalized patients undergoing total body irradiation (TBI) in the course of therapy. The classifier uses the same genes and gene weights for samples from any of the model systems and does not include interaction effects between gene and model system.
In addition to the generation of a biosignature, it was possible to use the gene-expression measurements to test for concordant differential expression in response to radiation among three model systems. While there is some evidence of concordant regulation of genes in the presence of radiation in the three systems, individual genes that are strongly associated with radiation exposure in all three systems are the exception rather than the rule. That being the case, it is concluded that, while it is believed that a biodosimeter has been constructed that will function in an otherwise healthy human population, it is also suspected that any such biodosimeter will be underperforming when compared to one that is trained on data generated from the population for which it is designed. Because such data are impossible to obtain, it is proposed that a full solution to the challenge of biodosimetry will involve a “best guess” biodosimeter based on available model systems together with a clear technique for incremental improvements in the field based on new training data as such data become available.
In summary, in order to obtain an exhaustive list of possible predictors of radiation dose response, three steps are iteratively repeated: 1) generate models from a candidate list of biomarkers using variable selection, 2) identify the genes in this model that account for 90% of model variability, and 3) flag these genes as important and remove them from the candidate list. This results in a collection of models, each with mildly lower accuracy than the previous. By visual inspection of each model, it is determined that accuracy falls off after model five, so all subsequent models are excluded. Plots of each of the top five models as well as the genes included in those models are included in
While the models perform well in all three systems (mouse, human ex vivo and human TBI), all of the data used to build these models is based on Affymetrix microarrays. As such, it is expected that these models would perform well in stratifying radiation exposure in other model systems if the data for those systems was generated by Affymetrix microarray. However, because of consistency and cost, these arrays may not be optimal candidates for a field device. The genes set forth in Table 9 (see also
indicates data missing or illegible when filed
All documents and other information sources cited above are hereby incorporated in their entirety by reference.
This application claims priority from U.S. Provisional Application No. 61/704,945, filed Sep. 24, 2012, the entire content of which is incorporated herein by reference.
This invention was made with government support under Grant Nos. Al-067798-01, AI-067798-06 and HL-086998-01, awarded by the National Institutes of Health. The government has certain rights in the invention.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US2013/000218 | 9/24/2013 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 61704945 | Sep 2012 | US |
