Next-nearest-neighbor sequence determinants of antisense DNA

Abstract

The use of antisense oligodeoxyribonucleotides (ODNs) to inhibit translation of mRNAs promises to be an important means of controlling gene expression and disease processes. ODNs are about 20 nucleotides long, so hundreds of possible targets are available in a given mRNA. An elusive goal has been to efficiently predict the best in vivo antisense target without having to study a large pool of possible ODN sequences for each mRNA. It would be a breakthrough if ODN selection could be accurately guided by the application of sequence specific parameters to an mRNA sequence. The selection of the best ODN sequence is complicated since cellular uptake, conditions at the mRNA target site, non-sequence-specific effects, sequence redundancy, and mRNA secondary structures are difficult to predict. Thermodynamic parameters for nearest-neighbor (dimer) duplex stabilities, from in vitro studies, have not been adequate predictors of in vivo hybridization. The methodology of this application shows that it is possible to obtain parameters for in vivo motifs, which are defined as combinations of next-nearest-neighbors, that are correlated with efficient antisense targeting. These parameters can be used to identify mRNA sequences that are binding sites for effective antisense ODNs. Next-nearest-neighbor nucleotide parameters can be derived directly from cell culture inhibition data so that in vivo conditions are taken into account.

Description

TECHNICAL FIELD OF INVENTION

[0003] The present invention relates generally to the field of antisense gene therapy and methods for identifying therapeutic oligodeoxyribonucleotides (ODNs) with the most suitable sequences for regulation of pathogenic processes associated with specific genetic diseases and for identifying oligodeoxyribonucleotides for the general control of gene expression, whether or not the gene is involved in a known genetic disorder.

BACKGROUND OF THE INVENTION

[0004] The field of antisense therapy involves techniques that attempt to treat a variety of disorders that are associated with genetic deficiencies or defects. One type of gene therapy treatment takes the form of treating the patient with a regulatory molecule, such as an antisense DNA oligodeoxyribonucleotide (ODN) molecule that binds to messenger RNA (mRNA) with the subsequent inhibition or control of translation and, hence, control of the production of a protein product. The antisense molecule is typically an oligonucleotide modified so as to have a long lifetime in the presence of cellular nucleases as well as to have high efficiency in hybridization to the target mRNA or genomic DNA. However, these modifications can result in undesirable side effects, one of which is that the ODN binds to cellular proteins and inhibits cellular functions in unpredictable ways [Mercola & Cohen, 1995; Orr & Monia 1998; Eckstein 2000]1. FIG. 1 illustrates the fact that the desired effect of an antisense ODN requires that the ODN reach the target mRNA in the cellular nucleus and that the ODN be able to selectively bind to a region (typically 20 nucleotides long) of the target mRNA, but not to other mRNAs. Non-specific binding to cellular proteins on the cell surface, in the cytoplasm, or in other compartments, including the nucleus, can reduce effective ODN concentrations. Therefore, the antisense effect in vivo is dependent on many factors. 1 All of the references cited herein and provided in the disclosed Bibliography are fully incorporated by reference in the specification.

[0005] Antisense oligodeoxyribonucleotides (ODNs), typically designed to be complementary to a specific mRNA target sequence of about 20 nucleotides, have been shown to be effective as a means of transient disruption of gene expression at the translational level [Sokol et al., 1998; Sokol & Gewirtz, 1999]. Thirteen antisense ODNs, six of which are targeted to cancer genes, are approved or are in clinical trials [Braasch & Corey, 2002]. The first generation of antisense drugs consists of phosphorothioate-modified oligodeoxyribonucleotides (S-ODNs), in which one of the non-bridging oxygens is replaced by sulfur to inhibit nuclease degradation. S-ODNs, like unmodified DNA, exert their effect mainly by activating RNAse H, which binds to the sites of S-ODN:mRNA hybridization and cleaves the mRNA. There is growing evidence that this antisense effect takes place in the nucleus, although the uptake mechanism and nuclear localization can depend on ODN concentration [Beltinger et al., 1995; Gray et al., 1997; Orr & Monia, 1998; Sokol & Gewirtz, 1999]. Methods are now available to correlate ODN:mRNA hybridization with a reduction in mRNA and protein levels [Sokol et al., 1998; Sokol & Gewitz, 1999].

[0006] S-ODNs can exert a true antisense inhibition of translation, which is sequence-specific, as exemplified in studies of C-raf and A-raf inhibition by S-ODNs with increasing numbers of mismatches [Coiffi et al., 1997]. However, a plethora of effects, broadly denoted as non-specific, compromise the ability to predict true sequence-specific antisense effects on the basis of in vitro hybridization data [Branch, 1998; Stein, 1999]. These non-specific effects include the competing secondary structures of mRNA target sites, partial complementarity of ODNs with unintended sites, the interactions of ODNs with intracellular and extracellular proteins, ODN self-structures such as G-quartet structures (although G-containing tetraplex structures may not form under intracellular conditions [Basu & Wickstrom, 1997]), effects of carriers, the cell type, the particular mRNA that is targeted, and conditions at the mRNA target site. Moreover, cellular delivery and subcellular trafficking may be somewhat sequence specific [Stein & Cheng, 1993; Wagner & Flanagan, 1997]. The type of ODN modification is also important. Chemical modifications other than phosphorothioate modification, such as 2′-O-alkyl and 2′-O-methoxyethoxy modifications, methylphosphonation, and 2′-5′ linkage of 3′-deoxyribonucleotides, have been used to increase the stability and reduce the non-specific effects of antisense S-ODNs [Monia et al., 1993; Gray et al., 1997; Giles et al., 1998]. However, these modifications reduce RNase H sensitivity. For this reason, chimeric antisense ODNs that combine such modifications together with five to seven simple phosphorothioate nucleotides (to retain RNase H sensitivity) have been advocated [Monia et al., 1993]. Phosphorothioate modification thus remains an important modification.

[0007] Non-specific effects are not necessarily bad if they offer an added source of drug potency [Branch, 1998]. However, non-specific effects of ODN sequences have been difficult to predict, and a bottleneck remaining in the design of antisense drugs is the inability to make rational, a priori, selections of the best mRNA target sequences [Branch, 1998; Bernstein, 1998; Eckstein, 1998]. Others in the field are moving toward streamlined testing of all possible accessible sites on a target mRNA [Eckstein, 1998; Ho et al., 1998; Matveeva et al., 1998].

SUMMARY OF THE INVENTION

[0008] The present invention proposes a different approach to take account of many non-specific effects and enhance the ability to rationally identify potential antisense target sites. The method of the present invention is to derive, directly from in vivo data, a sequence-dependent set of parameters that is correlated with the antisense effect and that can be used to aid selection of mRNA target sites. Previously, parameters from in vitro studies have been used. Because in vitro studies of oligomer duplexes have shown that thermodynamic stability can be described as a property of the nearest-neighbor (NN) base pairs, the simplest premise has been that NN stabilities might dominate the sequence-dependent antisense effect. All 16 distinguishable NN stabilities for DNA:RNA hybrids must be considered [Stull et al., 1992; Roberts & Crothers, 1992; Hung et al., 1994; Lesnik & Freier, 1995; Gray, 1997a; Gray, 1997b; Hashem et al., 1998; Ho et al., 1998]. For example, hybrids with adjacent purines in the RNA strand (e.g. r[AG]/d[CT]) are much more stable than those with adjacent pyrimidines in the RNA strand (e.g. r[CU]/d[AG]). Using published DNA:RNA hybrid stability data [Gray, 1997a; Gray, 1997b; Sugimoto et al., 1995], it is possible to calculate the theoretical variation in the free energy ΔG° (at 37° C., 1 M Na+) for hybridizations of ODNs to all target sequences in an mRNA. Differences in the stabilities of DNA:mRNA sequences, based on stabilities of the NN base pairs obtained from in vitro thermodynamic measurements of paired oligomer hybrids, have been widely used as an important factor in the design of antisense ODN sequences and in the assessment of differences in the antisense effectiveness of ODNs targeted to different regions of the same mRNA [Stull et al., 1992; Mathews et al., 1999; Gray & Clark, 1999; Gray & Clark, 2001; Walton et al., 2002]. The NN hybrid stabilities, while important, are generally supplemented in predictive programs with other criteria, such as the stability of competing intrastrand mRNA base pairs that must be broken to allow ODN:mRNA pairing [Mathews et al., 1999; Walton et al., 2002]. However, even then, such predictions are of limited accuracy, and methods based on in vitro criteria cannot account for sequence-dependent in vivo effects as described in the previous section.

[0009] Although NN data are not highly predictive of antisense effectiveness, this does not mean that data for longer sequences, or data obtained under cellular conditions, would also be poor predictors. In an important survey, Tu et al. (1998) found that fewer than 42 of 2026 reports involved testing more than 10 ODN sequences before concluding that there was an antisense effect. From a further analysis of 42 effective antisense sequences (including those used in clinical trials), Tu et al. (1998) discovered that 20 of these were targeted to a GGGA motif in the mRNA. In other work, Matveeva et al. (2000) reported that GUGG, GGGA, GAGU, UGGC, and AGAG target motifs were positively correlated with antisense effectiveness, while CCCC, CAGU, UUA, CCGG, and UUU were negatively correlated with antisense effectiveness in a search of >1000 experiments. The method provided by the present invention allows the identification of a more comprehensive, discrete number of motifs of overlapping triplet sequence combinations and provides an unambiguous way to rank all possible sequences in terms of the antisense effectiveness of the motifs they contain.

[0010] Specifically, the present invention provides a method for designating a nucleotide sequence composed of 20 nucleotides as 20 next-nearest-neighbor nucleotide triplets. Such method is comprised of the following operations: treating the nucleotide sequence composed of 20 nucleotides as a closed sequence, with the ends of said sequence meeting to form a circle, reading the sequence three nucleotides at a time, moving up one nucleotide along the sequence, then reading the next three nucleotides, for 20 steps, and interpreting the 20 readings of three nucleotides each as being equal to 20 next-nearest-neighbor nucleotide triplets.

[0011] The present invention also provides a method for designating a nucleotide sequence of n nucleotides as n next-nearest-neighbor nucleotide triplets. Such method comprises the following: treating the n nucleotide sequence as a closed sequence, with the ends of said sequence meeting to form a circle, reading the nucleotide sequence three nucleotides at a time, moving up one nucleotide along the sequence, then reading the next three nucleotides, for n steps, and interpreting the n readings of three nucleotides each as being equal to n next-nearest-neighbor nucleotide triplets.

[0012] Additionally, the present invention provides a method for assigning parameters of antisense effectiveness to 64 next-nearest-neighbor nucleotide triplets from measurements of the antisense effectiveness of at least 64 nucleotide sequences, wherein the sequences are considered to be closed sequences without end effects. This method of the present invention is comprised of operations including the step of constructing a matrix X having N rows and M columns, wherein said matrix X has one row for each of a minimum of N=64 nucleotide sequences and one column for each of the possible M=64 types of next-nearest-neighbor nucleotide triplets for sequences containing the common four nucleotides, A, U, G, and C, and wherein the numbers in the matrix columns are the numbers of each type of next-nearest-neighbor nucleotide triplet in the sequence in the given row. Said method further comprises the steps of constructing a matrix Y with N rows and 1 column, wherein the numbers in the rows are measured values for the biological antisense effectiveness of the N sequences, dividing the rows of matrices X and Y by the respective errors in the measured values, constructing a matrix P having M rows and 1 column, wherein the numbers in the rows are the 64 parameters of antisense effectiveness assigned to the next-nearest-neighbor nucleotide triplets, checking that the matrices satisfy the condition that X multiplied by P equals Y, via matrix multiplication, and solving equation X multiplied by P equals Y via the singular value decomposition method, wherein said equation is solved for P. In one embodiment of the present invention, the measurements of antisense effectiveness for this method are taken from a database of in vitro measurements. In another embodiment of the present invention, the measurements of antisense effectiveness for this method are from a database of in vivo measurements.

[0013] The method of the present invention for assigning parameters of antisense effectiveness to 64 next-nearest-neighbor nucleotide triplets from measurements of antisense effectiveness is further defined wherein the measurements of antisense effectiveness are taken from a database of antisense effects with sequences of phosphorothioate oligonucleotides. In a preferred embodiment of the present invention, such measurements of antisense effectiveness are taken from a database of antisense effectiveness with sequences of chemical moieties that pair with an mRNA sequence in complementary fashion. In yet another embodiment of the present invention, said parameters for said 64 next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of a phosphorothioate oligonucleotide of about 20 nucleotides in length. In one embodiment of the present invention, said parameters for said 64 next-nearest-neighbor nucleotide triplets, which are effective in determining the antisense effectiveness of a phosphorothioate oligonucleotide of about 20 nucleotides in length, are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

[0014] In the present invention, the parameters for said 64 next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of sequences of chemical moieties that pair with an mRNA sequence in complementary fashion for a length of about 20 nucleotides. The present invention further provides that said parameters for said 64 next-nearest-neighbor nucleotide triplets, which are effective in determining the antisense effectiveness of sequences of chemical moieties that pair with an mRNA sequence in complementary fashion for a length of about 20 nucleotides, are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

[0015] An embodiment of the present invention also exists wherein the parameters for 64 next-nearest-neighbor nucleotide triplets are combined to give parameters for 49 independent combinations of next-nearest-neighbor nucleotide triplets. Moreover, additional parameters are to be included to specify type of organism, type of cell line, type of gene mRNA, or type of chemically-modified oligomer.

[0016] The present invention further provides a method for assigning parameters to 49 combinations of next-nearest-neighbor nucleotide triplets from measurements of the antisense effectiveness of at least 49 sequences, wherein the sequences are considered to be closed sequences without end effects. Such method is comprised of operations including the constructing of a matrix X having N rows and M columns, wherein said matrix X has one row for each of a minimum of N=49 nucleotide sequences and there is one column for each of the possible M=49 independent combinations of next-nearest-neighbor nucleotide triplets for sequences containing the common four nucleotides A, U, G, and C, wherein the numbers in the matrix columns are the numbers of each type of independent next-nearest-neighbor nucleotide combination in the sequence in the given row. Said method further comprises the operations of constructing a matrix Y having N rows and 1 column, wherein the numbers in the rows are the measured values for the biological antisense effectiveness of the N sequences, dividing the rows of matrices X and Y by the respective errors in the measured values, constructing a matrix P having M rows and 1 column, wherein the numbers in the rows are the 49 parameters of antisense effectiveness assigned to the independent next-nearest-neighbor nucleotide combinations, checking that the matrices satisfy the condition that X multiplied by P equals Y, via matrix multiplication, and solving the equation X multiplied by P equals Y via the singular value decomposition method, wherein said equation is solved for P.

[0017] In a preferred embodiment of the present invention, the measurements of antisense effectiveness for this method are taken from a database of in vitro measurements. In another preferred embodiment of the invention, the measurements of antisense effectiveness for this method are taken from a database of in vivo measurements. Said method provides that said measurements of antisense effectiveness are taken from a database of antisense effects with sequences of phosphorothioate oligonucleotides. In addition, the present invention provides that said measurements of antisense effectiveness are taken from a database of antisense effectiveness with sequences of chemical moieties that pair with an mRNA sequence in complementary fashion. The invention further provides that the parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of a phosphorothioate oligonucleotide of about 20 nucleotides in length, wherein said parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

[0018] Under the method of the present invention, the parameters for the 49 combinations of next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of sequences of chemical moieties that pair with an mRNA sequence in complementary fashion for a length of about 20 nucleotides, wherein said parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are multiplied by L/20 for an oligonucleotide which is L nucleotides long. The invention provides that, in assigning parameters to 49 combinations of next-nearest-neighbor nucleotide triplets, additional parameters are to be included to specify type of organism, type of cell line, type of gene mRNA, or type of chemically-modified oligomer.

[0019] Sequence motifs longer than the nearest-neighbors have not been searched for in a comprehensive manner, and they have never been rigorously searched for using in vivo databases. The present invention provides a method for (1) identifying next-nearest-neighbor (NNN) triplet combinations as an objective description of motifs that describe antisense effectiveness and (2) using next-nearest-neighbor parameters to identify antisense targets on mRNAs. This method takes advantage of the fact that many of the factors that determine in vivo sequence-dependence antisense effectiveness are similar for various situations. Therefore, the detailed description that follows below is based on a limited set of data. From the following description, it will be recognized by those skilled in the art that NNN combinations can eventually be derived from a larger database that will allow discovery and comparisons of antisense-affiliated motifs in different genes, in different regions (coding, non-coding) of genes, for different cell lines, and for different ODN modifications.

BRIEF DESCRIPTION OF THE DRAWINGS

[0020] FIG. 1. A schematic illustration depicting antisense control of gene expression. To achieve an inhibitory effect on the expression of a particular gene product, the antisense ODN sequence must reach the cellular nucleus and bind to a complementary sequence on the mRNA transcript, whereby the mRNA is cleaved by an endogenous RNAse H enzyme. The cleaved mRNA is thus rendered ineffective as a template for protein synthesis in the cytoplasm.

[0021] FIGS. 2A, 2B, and 2C. Illustrations of a 3×64 matrix showing how three 20-mer sequences can be represented in terms of their next-nearest-neighbor triplet compositions. Each sequence is considered to be a closed circle, so that it has 20 NNN triplets.

[0022] FIG. 3. Graphical illustration of inhibition of protein levels of four gene products in 102 antisense DNA experiments versus (A) levels predicted by in vitro nearest-neighbor parameters and (B) levels predicted by in vivo next-nearest-neighbor parameters. Error bars are ± one standard deviation.

[0023] FIG. 4. Graphical illustration of inhibition of PKCα protein levels in 20 antisense DNA experiments (Dean et al., 1994); Western blot assay for loss in protein level versus (A) levels predicted by in vitro nearest-neighbor parameters and (B) levels predicted by in vivo next-nearest-neighbor parameters. Error bars are ± one standard deviation.

[0024] FIG. 5. Graphical illustration of inhibition of adhesion molecule mRNA levels in 33 antisense DNA experiments (Bennett et al., 1994; Northern blot assay for loss in mRNA level) versus (A) levels predicted by in vitro nearest-neighbor parameters and (B) levels predicted by in vivo next-nearest-neighbor parameters. Error bars are ± one standard deviation.

[0025] FIG. 6. Graphical illustration of inhibition of P-glycoprotein function in 22 antisense DNA experiments (Ho et al., 1996; rhodamine flux assay) versus (A) levels predicted by in vitro nearest-neighbor parameters and (B) levels predicted by in vivo next-nearest-neighbor parameters. Error bars are ± one standard deviation.

DETAILED DESCRIPTION OF THE INVENTION

[0026] The present invention involves the following operations:

[0027] (1) A database of antisense effects must be obtained for a number of sequences that is greater than the number of next-nearest-neighbor parameters that need to be determined. The database must have at least one example of each NNN triplet in at least one sequence. The method of the present invention is unique in that the database may consist of data obtained for in vivo antisense effects, but those skilled in the art will realize that other databases may be used to determine NNN parameters.

[0028] From our published theory [Gray, 1997a; Gray 1997b], if end effects are significant, 84 NNN parameters are needed to describe an array of sequence-dependent data for short oligomers. In the case of relatively long ODNs of 20-nucleotides, one can treat the sequences as closed sequences with no ends, which reduces the number of independent parameters to 49 [Gray, 1997a; Gray 1997b]. In this latter case, there are fewer than 64 (=43) NNN parameters because there are 15 constraints on arranging triplets in a closed sequence. A search of the literature and 42 references from Tu et al. (1998), revealed no single database large enough for such an analysis. In addition, only a few data sets of 20 or more included individual errors, which are needed to weight the % inhibition values for a singular value decomposition (SVD) analysis. In the present invention, data were obtained for a total of 102 antisense ODN sequences that contained representative numbers of all 64 NNN triplets. The ODNs were uniformly modified to contain phosphorothioate linkages between each nucleotide, with no phosphate groups at either end. Four gene products were targeted: C-Rafl, AKT2, Bcl-2, and PKCα, and the data was obtained from antisense treatments of two cell lines, T24 bladder cancer cells and A549 lung cancer cells. The specific sequence positions on the mRNAs and the inhibition data of protein levels are given in Table 1.

TABLE 1
Inhibition of protein levels from four genes when treated with 76
different phosophorothioate ODNs. Since 26 of the ODNs were used
to treat two different cell lines for C-Raf1 inhibition, the
total number of data points was 102.*
	Sequence starting	% Inhibition of		% Inhibition of
	position in mRNA of	protein level in		protein level in
Gene	20-mer target	T24 cell line	Error (%)	A549 cell line	Error (%)
C-Rail	1	36.7	9.7	26.4	***
	41	46.8	4.6	47.7	***
	61	53.3	13.5	29.9	***
	85	39	15.9	40.8	9.70
	121	53.3	1.3	50.1	***
	130	32.9	11.5	42.9	9.80
	181	37.1	2.9	45.3	***
	301	34	***	31.9	***
	361	39	***	38.2	***
	707	34.5	12.6	22.1	10.90
	761	29.4	6.6	44.5	9.80
	821	43.1	7.4	51.2	10.10
	1041	36.2	11.3	48.3	0.90
	1063	47.6	18.5	48.6	10.70
	1181	57.1	0.3	54.3	10.10
	1474	48	12.3	51.7	7.40
	1777	39.7	13.2	51.7	10.80
	1867	42.5	10.5	55.5	5.50
	2098	45	12.2	45.8	12.40
	2341	66.1	25.2	59.9	0.30
	2349	35.9	10.8	36.9	11.30
	2484	59.2	20.1	52.4	4.80
	2581	56.7	14.5	63	2.80
	2601	24	2.4	42.1	12.70
	2661	42.3	8.7	66.3	0.00
	2681	42	1.8	49.2	4.10
AKT2	57			41	14.00
	86			54	5.00
	94			46	6.00
	490			58	7.00
	973			55	9.00
	1188			28	13.00
	1227			41	17.00
Bcl-2	23			41	14.00
	36			54	5.00
	61			46	6.00
	109			58	7.00
	120			55	9.00
	201			28	13.00
	277			41	17.00
	316			51	6.00
	361			30	14.00
	409			46	8.00
	445			53	8.00
	453			45	3.00
	501			35	9.00
	541			34	12.00
	601			28	11.00
	1041			50	5.00
	1421			53	8.00
	1481			56	5.00
	1641			49	10.00
	1761			52	5.00
	1821			28	10.00
	1941			50	10.00
	2081			43	10.00
	3101			37	6.00
	3921			41	6.00
	3941			43	4.00
	3961			42	6.00
	4881			44	7.00
	5321			45	4.00
PKCα	121	42.0	9.0
	281	49.5	9.8
	288	51.0	1.0
	301	22.0	9.0
	321	24.0	2.0
	341	29.0	4.0
	421	16.7	17.6
	441	18.6	10.2
	481	11.0	3.0
	501	53.0	7.0
	541	15.0	1.0
	621	22.8	11.3
	899	62.6	3.9
	2044	51.3	6.8
*From a preliminary NNN fit by SVD, the % inhibition data for five additional S-ODNs targeted to C-Rafl mRNA had squared deviations of twice the average and were omitted from the final set.
**Errors are ranges from duplicate Western blots or standard deviations from three or more measurements
***Where no error is shown, the data are from single measurements and for the purpose of SVD analysis a maximum error of 15% was assumed.

[0029] (2) Each of the target mRNA sequences is separated into its constituent NNN triplets. This is illustrated in FIGS. 2A, 2B, and 2C for three of the target sequences in C-Rafl mRNA. The total number of NNN triplets is 20 for each sequence, but, in general, they are different for each sequence. The array shown in FIGS. 2A, 2B, and 2C makes a 3×64 matrix. An additional row is added for each additional sequence. Additional columns in the matrix may be added to allow for differences in the data sets for different genes or cell lines or any other parameter than one wants to distinguish. For example, we added a 65th column with a “0” for every sequence that was used to inhibit the level of a gene product in T24 cells and a “1” for every sequence that was used to inhibit the level of a gene product in A549 cells. For the sequences in Table 1, the resulting matrix was 102 (# of sequences)×65 (# of parameters).

[0030] (3) Solve the matrix equation. The matrix equation NNNhk×Pk=Ih, where the NNN matrix has dimensions of h=102 and k=65, the P vector has k=65 values, one for each of the 64 NNN triplexes and one for the cell line, and I is the vector of % inhibition values for each sequence. Each hth row of the NNNhk matrix and the hth value of the Ih vector is divided by the error for the hth sequence. This equation was solved as in our other work (Gray 1997b) using standard procedures (Press et al., 1992) to give values for the 65 parameters of the P vector and, hence, for the 49 independent combinations of the NNN triplets. The values for these parameters are listed in Table 2 for our specific data set. Those skilled in the art will realize that other sets of P parameters may be obtained for other data sets by the same procedure.

TABLE 2
Solution for the parameters P obtained from an SVD
solution to fit the inhibitory data for 102
antisense sequences targeted to four gene
mRNAs in two cell lines.
	Antisense inhibitory
Next-Nearest-	parameter from fit to	Number of NNN
Neighbor Triplet	102 sequences	Triplet in Data Set
AAA	8.323	25
AAU	−6.973	27
MG	6.819	24
AAG	0.455	33
AUA	−4.545	15
AUU	1.258	20
AUG	1.728	16
AUG	0.872	64
ACA	−5.736	41
ACU	3.613	28
AGO	5.101	21
AGG	4.638	25
AGA	2.685	36
AGU	−3.902	11
AGO	0.196	38
AGG	8.428	50
UAA	−3.399	6
UAU	−6.522	16
UAG	1.179	15
UAG	11.364	8
UUA	−1.557	9
UUU	−3.087	22
UUG	0.294	22
UUG	6.464	27
UGA	3.546	29
UGU	−0.072	15
UGG	6.405	32
UGG	−0.502	14
UGA	−2.802	35
UGU	−1.054	30
UGG	9.247	45
UGG	−1.498	59
GAA	2.841	31
GAU	6.933	28
GAG	−1.620	45
GAG	−4.277	35
GUA	16.112	10
GUU	−2.003	22
GUG	1.603	30
GUG	−1.000	41
OCA	−3.765	24
CCU	5.983	31
CCC	−4.530	29
CCG	6.198	54
CGA	5.110	34
CGU	−2.940	18
CGC	7.598	32
CGG	5.037	49
GAA	0.859	47
GAU	5.875	44
GAC	1.238	31
GAG	−0.134	59
GUA	−7.389	11
GUU	5.945	16
GUC	5.753	22
GUG	−2.444	37
GCA	9.832	45
GCU	5.189	29
GCC	−3.090	56
GCG	4.470	40
GGA	2.845	76
GGU	9.763	27
GGC	−0.641	55
GGG	−0.194	74

[0031] The P parameters show the significance of various NNN triplets to the antisense inhibitory effect of the sequences in the data set. Negative values mean that some triplets, or combinations of triplets, are actually counterproductive to a maximum antisense effect. One may also note that from the last column in Table 2 that the number of occurrences of the various triplets ranged from 6 to 76 in the sequences used in this database, so all NNN triplets were well represented.

[0032] (4) Assess the importance of combinations of the NNN triplets. Table 3 shows the relative importance of the 10 simplest independent combinations of NNN in the target sequences that were analyzed. Because there are constraints linking the NNN, only values for the four triplets that are homopurine or homopyrimidine can be individually determined (three left-hand columns of Table 3). The other 60 triplet values are interrelated and must be expressed as combinations, the simplest of which constitute six repeating sequences ((CG)n, (AC)n, etc.), and these are listed in the three right-hand columns of Table 3. The values in Table 3 reveal that: (a) triplets of RNA purines (GGG, AAA, and AGA+GAG) are all more important than those with RNA pyrimidines (CCC, UUU, and UCU+CUC), consistent with in vitro hybrid oligomer stabilities; (b) GGG is one of the most stable triplets, in agreement with Tu's analysis; (c) AAA also has an unusually high positive effect when it is present in antisense ODNs; and (d) UUU has a negative contribution, in agreement with the well-known instability of rU:dA pairs, which plays an important role in transcription termination in prokaryotes. Overall, the results of this analysis show that such an approach can give parameters pertinent to S-ODN:RNA hybridization in vivo and that results from more extensive data sets will lead to new insights regarding mRNA target selection.

TABLE 3
Values for 10 independent NNN triplets and independent
combinations of NNN triplets from an SVD solution to fit
the inhibitory data for 102 antisense sequences targeted
to four gene mRNAs in two cell lines.
	Antisense	Number of	Six Independent	Antisense	Avg Number of
Independent	Inhibitory	NNN Triplet	Combinations of	Inhibitory	Two NNN
Triplets	Parameter	in Data Set	Two NNN Triplets	Parameter	Triplets
AAA	8.323	25	CGC + GCG	6.034	36.0
GGG	−0.194	74	AGA + GAG	1.275	47.5
UUU	−3.087	22	UCU + CUC	0.765	22.5
CCC	−4.530	29	UGU + GUG	−1.749	33.5
			ACA + CAC	−3.678	43.0
			AUA + UAU	−5.534	15.5

[0033] (5) Apply the derived NNN values to the prediction of antisense effectiveness of other targeted genes. This procedure simply involves the multiplication of each NNN parameter from Table 2 with the number of occurrence of that NNN in the sequence for which a predicted value is desired. If the NNN values have been derived from sequences that are 20 nucleotides long and a predicted value is desired for sequences that are L nucleotides long where L is not 20, the NNN values should be multiplied by L/20. Examples are in the following section.

WORKING EXAMPLES

Example A

[0034] FIG. 3A shows the fit of hybrid NN free energy values, ΔG° (37° C.), typical of NN stability predictions, to the measured inhibition values for the 102 antisense ODN sequences used to inhibit four gene products in two cell lines (data from Table 1). FIG. 3B shows the fit of the derived NNN values in Table 2, multiplied by the NNN triplets in each of the 102 ODN sequences, to the measured inhibition values. The fit to the NNN data set is better, as is shown by the correlation coefficients, r, and the significance values from the t-text, P, in the first row in Table 4. That is, the NN fit has a regression coefficient of 0.309, while the NNN fit gives a better regression coefficient of 0.778, and a lower value of P, although both fits are significant (below the P=0.05 level).

TABLE 4
Comparison of the regression coefficients, r, and the t-test P
significance values from fits of various data sets with NN free
energy values and with NNN parameters (Table 2).
	Fit with Nearest-Neighbor	Fit with Next-Nearest-
Measured data and	Parameters*	Neighbor Parameters*
calculation being	Number
compared	of sequences	rNN	PNN	rNNN	PNNN
102 sequences from	102	0.309	>0.002	0.778	<0.0001
Table 1
Published PKC	20	0.368	0.110	0.536	0.015
alpha data (Dean et
al., 1994)
Published adhesion	33	0.134	0.457	0.267	0.133
molecules (Bennett
et al., 1994)
Published P-	22	0.879	<0.0001	0.396	0.068
glycoprotein data
(Ho et al., 1996)
*r is the coefficient of correlation and P is the significance of r using the t-test. The smaller the value of P the more significant the correlation; P is the level at which the null hypothesis can be rejected.

Example B

[0035] In this example, the predictions from the NNN parameters are compared with those from the NN free energy ΔG° for a data set that was not used in deriving the NNN parameters of Table 2. The data are published inhibition data from Western blots of PKCα protein taken after treatments with 20 antisense S-ODNs (Dean et al., 1994). As shown in FIGS. 4A and 4B, and in the second line of data in Table 4, the experimental inhibition data are better approximated by those of the in vivo NNN values than by the in vitro NN parameters. It is noteworthy that the NNN parameters give a fit that is significant (P=0.015), while the NN parameters give a fit that is not significant at the P=0.05 level (P=0.110).

Example C

[0036] In this example, the predictions from the NNN parameters are compared with those from the NN free energy ΔG° for a second data set that was not used in deriving the NNN parameters of Table 2. The data are published inhibition data from Northern blots for two adhesion molecule mRNAs (E-Selectin and ICAM-1) taken after treatments with 33 antisense S-ODNs (Bennett et al., 1994). As shown in FIGS. 5A and 5B, and in the third line of data in Table 4, this experimental inhibition data is also better approximated by those of the in vivo NNN values than by the in vitro NN parameters. However, in neither case is the P value below the desired 0.05 value, indicating that additional factors remain to be identified by using a larger data set.

Example D

[0037] In this example, the predictions from the NNN parameters are compared with those from the NN free energy ΔG° for a fourth data set that was not used in deriving the NNN parameters of Table 2. The data are published inhibition data from inhibition of P-glycoprotein function in 22 antisense DNA experiments using a rhodamine flux assays (Ho et al., 1996). In this case, as may be seen in FIGS. 5A and 5B, and in the last line of data in Table 4, the experimental inhibition data are better approximated by those of the in vitro NN parameters (P<0.05), although the in vivo NNN parameters give a fit with P=0.068, close to the desired P=0.05 level. The ODN sequences used in this study were preselected by an in vitro library of oligonucleotides to test for RNase H sensitive sites. This case serves to illustrate that the NNN methodology differs from methods based on NN stabilities of DNA:mRNA hybrids and is sensitive to different factors during in vivo antisense treatments.

COMPARISON WITH OTHER METHODOLOGIES

[0038] Prior art in identifying effective antisense sequence combinations may be divided into two categories. In the first category, the majority of predictive routines are based on the knowledge that the thermodynamic properties of nucleic acid duplexes reside in the interactions of neighboring bases or base pairs, called nearest-neighbor (NN) properties. The predictive algorithms provided in computer programs including Gray & Clark, 1999; Gray & Clark, 2001, HYBsimulator™ software (RNAture, Inc; Mitsuhashi et al., 1994), OligoWalk (Mathews et al., 1999), a recent program by Walton et al. (2002) and all similar programs, to the knowledge of the inventor, rely on NN properties of nucleic acid duplexes, sometimes augmented with other factors such as changes in the folding of the target mRNA sequence. There are only 13 independent combinations of NN properties in closed sequences. The present invention differs in that it relies on NNN triplet properties of sequences. There are 49 independent NNN triplet combinations for closed sequences. In the second category, researchers have searched for motifs of three or more nucleotides that are present in effective antisense molecules (Tu et al., 1998; Matveeva et al., 2000). The results of such searches do not, however, show how to optimize motif combinations in a given target sequence. Nor do such results allow one to objectively rank the antisense effectiveness of all sequences regardless of differences in their motif combinations. In fact, the theory of nucleotide sequence combinations (Gray 1997a; 1997b) shows that motifs cannot be combined in all combinations. The present invention allows for the derivation of a minimal set of 64 parameters for 64 NNN triplets that can be used to assign parameters for up to 49 independent NNN triplet combinations (i.e. motifs). Values such as those in Table 2 above are sufficient to account for all NNN triplet combinations (i.e. motifs) needed to unambiguously rank the effectiveness of any nucleotide sequence, as long as the sequence is long enough to be considered a closed, circular sequence.

[0039] To those knowledgeable in the art, the present method may be expanded to derive 49 NNN combinations from any database with at least 49 sequences for closed sequences. The method may be expanded to cover sequences that are not closed and to derive next-next-nearest-neighbor (NNNN) quadruplet properties.

BIBLIOGRAPHY

[0040] The following references are hereby specifically incorporated herein by reference:

[0041] Basu, S., & E. Wickstrom (1997) Nucl. Acids Res. 25, 1327-1332.

[0042] Beltinger, C., H. Saragovi, R. Smith, L LeSauteur, N. Shah, L. DeDionisio, L. Christensen, A. Raible, L. Jarett, & A. Gewirtz (1995) J. Clin. Invest. 95, 1814-1823.

[0043] Bennett, C. F., T. P. Condon, S. Grimm, H. Chan, & M.-Y. Chiang (1994) J. Immun. 152, 3530-3540.

[0044] Bernstein, P. (1998) Nature Biotechnology “In Brief” 16, 10.

[0045] Braasch, D. A., & D. R. Corey (2002) Biochemistry 41, 4503-4510.

[0046] Branch, A. D. (1998) Trends Biochem. Sci. 23, 45-50.

[0047] Cioffi, C. L., M. Garay, J. F. Johnston, K. McGraw, R. T. Boggs, D. Hreniuk, & B. P. Monia (1997) Mol. Pharm. 51, 383-389.

[0048] Dean, N. M., R. McKay, T. P. Condon, & C. F. Bennett (1994) J. Biol. Chem. 269, 16416-16424.

[0049] Eckstein, F. (1998) Nature Biotechnology “In Brief” 16, 24.

[0050] Eckstein, F. (2000) Antisense & Nucleic Acid Drug Development 10, 117-121.

[0051] Giles, R. V., D. G. Spiller, J. Grzybowski, R. E. Clark, P. Nicklin, & D. M. Todd (1998) Nucl. Acids Res. 26, 1567-1575.

[0052] Gray, D. M. (1997a) Biopolymers, 42, 783-793.

[0053] Gray, D. M. (1997b) Biopolymers, 42, 795-810.

[0054] Gray, D. M., and C. L. Clark (1999) “Method for Selectively Ranking Sequences for Antisense Targeting” U.S. Pat. No. 5,856,103 issued Jan. 5, 1999.

[0055] Gray, D. M., and C. L. Clark (2001) “An Apparatus and Method for Selectively Ranking Sequences for Antisense Targeting” U.S. Pat. No. 6,183,966B1 issued Feb. 6, 2001.

[0056] Gray, G. D., S. Basu, & E. Wickstrom (1997) Biochem. Pharm. 53, 1465-1476.

[0057] Hashem, G. H., J. Pham, M. R. Vaughan, & D. M. Gray (1998) Biochemistry, 37, 61-72.

[0058] Ho, S. P., D. H. O. Britton, B. A. Stone, D. L. Behrens, L. M. Leffet, F. W. Hobbs, J. A. Miller, & G. L. Trainor (1996) Nucl. Acids Res. 24, 1901-1907.

[0059] Ho, S. P., Y. Bao, T. Lesher, R. Malhotra, L. Y. Ma., S. J. Fluharty, & R. R. Sakai (1998) Nat. Biotech. 16, 59-63.

[0060] Hung, S.-H., Q. Yu, D. M. Gray, & R. L. Ratliff (1994) Nucl. Acids Res. 22, 4326-4334.

[0061] Lesnik, E. A., & S. M. Freier (1995) Biochemistry 34, 10807-10815.

[0062] Mathews, D. H., M. E. Burkard, S. M. Freier, J. R. Wyatt, & D. H. Turner (1999) RNA 5, 1458-1569.

[0063] Matveeva, O., B. Felden, A. Tsodikov, J. Johnston, B. P. Monia, J. F. Atkins, R. F. Gesteland, & S. M. Freier (1998) Nat. Biotech. 16, 1374-1375.

[0064] Matveeva, O. V., A. D. Tsodikov, M. Giddings, S. M. Freier, J. R. Wyatt, A. N. Spiridonov, S. A. Shabalina, R. F. Gesteland, & J. F. Atkins (2000) Nucl. Acids Res 28, 2862-2865.

[0065] Mercola, D, & Cohen J. S. (1995) Cancer Gene Ther. 2, 47-59.

[0066] Mitsuhashi, M., Cooper, A., Ogura, M., Shinagawa, T., Yano, K., & Hosokawa, T. (1994) Nature 367, 759-761.

[0067] Monia, B. P., E. A. Lesnik, C. Gonzalez, W. F. Lima, D. McGee, C. J. Guinosso, A. M. Kawasaki, P. D. Cook, & S. M. Freier (1993) J. Biol. Chem. 268, 14514-14522.

[0068] Orr, R. M., & B. P. Monia (1998) Curr. Res. Mol. Therapeutics 1, 102-108.

[0069] Press, W. H., S. A. Teukolsky, W. T. Vetterling, & B. P. Flannery (1992) Numerical recipes in C, 2nd ed., Cambridge University Press, NY, pp. 59-70.

[0070] Roberts, R. W., & D. M. Crothers (1992) Science 258, 1463-1466.

[0071] Sokol, D. L, X. Zhang, P. Lu, A. M. Gewirtz (1998) Proc. Natl. Acad. Sci. USA 95, 11538-11543.

[0072] Sokol, D. L., & A. M. Gewirtz (1999) Methods: Companion Meth. Enzymol. 17, 219-230.

[0073] Stein, C., & Y. Cheng (1993) Science 261, 1004-1012.

[0074] Stein, C. A. (1999) Nat. Biotech. 17, 209.

[0075] Stull, R. A., L. A. Taylor, & F. C. Szoka, Jr. (1992) Nucl. Acids Res. 20, 3501-3508.

[0076] Sugimoto, N., S. Nakano, M. Katoh, A. Matsumura, H. Nakamuta, T. Ohmichi, M. Yoneyama, & M. Sasaki (1995) Biochemistry 34, 11211-11216.

[0077] Tu, G.-C., Q.-N. Cao, F. Zhou, & Y. Israel (1998) J. Biol. Chem. 273, 25125-25131.

[0078] Wagner, R., & W. Flanagan (1997) Mol. Med. Today 3, 31-38.

[0079] Walton, S. P., G. N. Stephanopoulos, M. L. Yarmush, & C. M. Roth (2002) Biophysical J. 82, 366-377.

Claims

1. A method for designating a nucleotide sequence composed of 20 nucleotides as 20 next-nearest-neighbor nucleotide triplets comprising: (a) treating the nucleotide sequence composed of 20 nucleotides as a closed sequence, with the ends of said sequence meeting to form a circle; (b) reading the sequence three nucleotides at a time, moving up one nucleotide along the sequence, and then reading the next three nucleotides, for 20 steps; and (c) interpreting the 20 readings of three nucleotides each as being equal to 20 next-nearest-neighbor nucleotide triplets.

2. A method for designating a nucleotide sequence of n nucleotides as n next-nearest-neighbor nucleotide triplets comprising: (a) treating the n nucleotide sequence as a closed sequence, with the ends of said sequence meeting to form a circle; (b) reading the nucleotide sequence three nucleotides at a time, moving up one nucleotide along the sequence, and then reading the next three nucleotides, for n steps; and (c) interpreting the n readings of three nucleotides each as being equal to n next-nearest-neighbor nucleotide triplets.

3. A method for assigning parameters of antisense effectiveness to 64 next-nearest-neighbor nucleotide triplets from measurements of the antisense effectiveness of at least 64 nucleotide sequences, wherein the sequences are considered to be closed sequences without end effects comprising: (a) constructing a matrix X having N rows and M columns, wherein said matrix X has one row for each of a minimum of N=64 nucleotide sequences and one column for each of the possible M=64 types of next-nearest-neighbor nucleotide triplets for sequences containing the common four nucleotides, A, U, G, and C, wherein the numbers in the matrix columns are the numbers of each type of next-nearest-neighbor nucleotide triplet in the sequence in the given row; (b) constructing a matrix Y with N rows and 1 column, wherein the numbers in the rows are measured values for the biological antisense effectiveness of the N sequences; (c) dividing the rows of matrices X and Y by the respective errors in the measured values; (d) constructing a matrix P having M rows and 1 column, wherein the numbers in the rows are the 64 parameters of antisense effectiveness assigned to the next-nearest-neighbor nucleotide triplets; (e) checking that the matrices satisfy the condition that X multiplied by P equals Y, via matrix multiplication; and (f) solving equation X multiplied by P equals Y via the singular value decomposition method, wherein said equation is solved for P.

4. The method of claim 3, wherein said measurements of antisense effectiveness are taken from a database of in vitro measurements.

5. The method of claim 3, wherein said measurements of antisense effectiveness are taken from a database of in vivo measurements.

6. The method of claim 3, wherein said measurements of antisense effectiveness are taken from a database of antisense effects with sequences of phosphorothioate oligonucleotides.

7. method of claim 3, wherein said measurements of antisense effectiveness are taken from a database of antisense effectiveness with sequences of chemical moieties that pair with an mRNA sequence in complementary fashion.

8. The method of claim 3, wherein said parameters for said 64 next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of a phosphorothioate oligonucleotide of about 20 nucleotides in length.

9. The method of claim 8, wherein said parameters for said 64 next-nearest-neighbor nucleotide triplets are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

10. The method of claim 3, wherein said parameters for said 64 next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of sequences of chemical moieties that pair with an mRNA sequence in complementary fashion for a length of about 20 nucleotides.

11. The method of claim 10, wherein said parameters for said 64 next-nearest-neighbor nucleotide triplets are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

12. The method of claim 3, wherein said parameters for said 64 next-nearest-neighbor nucleotide triplets are combined to give parameters for 49 independent combinations of next-nearest-neighbor nucleotide triplets.

13. The method of claim 3, wherein additional parameters are to be included to specify type of organism, type of cell line, type of gene mRNA, or type of chemically-modified oligomer.

14. A method for assigning parameters to 49 combinations of next-nearest-neighbor nucleotide triplets from measurements of antisense effectiveness of at least 49 sequences, wherein the sequences are considered to be closed sequences without end effects comprising: (a) constructing a matrix X having N rows and M columns, wherein said matrix X has one row for each of a minimum of N=49 nucleotide sequences and there is one column for each of the possible M=49 independent combinations of next-nearest-neighbor nucleotide triplets for sequences containing the common four nucleotides A, U, G, and C, wherein the numbers in the matrix columns are the numbers of each type of independent next-nearest-neighbor nucleotide combination in the sequence in the given row; (b) constructing a matrix Y having N rows and 1 column, wherein the numbers in the rows are the measured values for the biological antisense effectiveness of the N sequences; (c) dividing the rows of matrices X and Y by the respective errors in the measured values; (d) constructing a matrix P having M rows and 1 column, wherein the numbers in the rows are the 49 parameters of antisense effectiveness assigned to the independent next-nearest-neighbor nucleotide combinations; (e) checking that the matrices satisfy the condition that X multiplied by P equals Y, via matrix multiplication; and (f) solving the equation X multiplied by P equals Y via the singular value decomposition method, wherein said equation is solved for P.

15. The method of claim 14, wherein said measurements of antisense effectiveness are taken from a database of in vitro measurements.

16. The method of claim 14, wherein said measurements of antisense effectiveness are taken from a database of in vivo measurements.

17. The method of claim 14, wherein said measurements of antisense effectiveness are taken from a database of antisense effects with sequences of phosphorothioate oligonucleotides.

18. The method of claim 14, wherein said measurements of antisense effectiveness are taken from a database of antisense effectiveness with sequences of chemical moieties that pair with an mRNA sequence in complementary fashion.

19. The method of claim 14, wherein said parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of a phosphorothioate oligonucleotide of about 20 nucleotides in length.

20. The method of claim 19, wherein said parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

21. The method of claim 14, wherein said parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are effective in determining the antisense effectiveness of sequences of chemical moieties that pair with an mRNA sequence in complementary fashion for a length of about 20 nucleotides.

22. The method of claim 21, wherein said parameters for said 49 combinations of next-nearest-neighbor nucleotide triplets are multiplied by L/20 for an oligonucleotide which is L nucleotides long.

23. The method of claim 14, wherein additional parameters are to be included to specify type of organism, type of cell line, type of gene mRNA, or type of chemically-modified oligomer.