Journal of Acquired Immune Dej’icicienc~ Syadmmes and Human Retrovirology 15:356-362 0 1997 Lippincott-Raven Publishers, Philadelphia Application of an Expert System in the Management HIV-Infected Patients *Michael J. Pazzani, tDarry1 See, j-Edison Schroeder, and iJeremiah Tilles of Departments of *Information and Computer Sciences and :Medicine, University of California, Irvine, Orange, California, U.S.A Summary: A rule-based expert system, Customized Treatment Strategies for HIV (CTSHIV), which encodes information from the literature on known drug-resistant mutations was developed. Additional rules include ranking and weighting based on antiviral activities, redundant mechanisms of action, overlapping toxicities, relative levels of drug-resistance, and proportion of drug-resistant clones in the HIV quasispe- ties. Plasma was obtained from HIV-infected patients and the RNA was extracted. Segments of the HIV pol gene encoding the entire protease, reverse transcriptase, and integrase proteins were amplified by reverse transcriptase-polymerase chain reaction (using a total of three primer pairs) and cloned. Sequencing was performed on five clones from each of two patients. When the patient’s RNA sequencing data were entered into the expert program, and the information was downloaded directly into the CTSHIV program, the five most effective two, three, and four drug regimens coupled with an explanation for their choice were displayed for each patient. Thus, the CT- SHIV system couples efficient genetic sequencing with an expert program that rec- ommends regimens based on information in the current medical literature. It may serve as a useful tool in the design of clinical trials and in the management of HIV-infected patients. Key Words: AIDS-Expert program-PCR-Cloning-Sequencing- Antiretrovirals. Many compounds have been reported to inhibit repli- cation of the human immunodeficiency virus (HIV) both in vitro and in vivo (1:2). In fact, there are already 11 antiretroviral agents licensed for use in the United States, and many more are currently undergoing preclinical and clinical evaluation. The approved agents include 2’,3’ dideoxynucleotide analog, reverse transcriptase inhibi- tors (azidothymidine [AZT], didanosine, dideoxycyti- dine, starvudine [d4T], and lamivudine [3TC]); non- nucleoside reverse transcriptase inhibitors (NNRTIs; Nevirapine, Delavirdine) and protease inhibitors (Sa- quinavir, Ritonavir, Indinavir, and Nelfinavir). Multiple clinical trials have demonstrated that treatment with these agents can result in improvement in markers such as CD4 count and viral load (3-5). Other studies have Address correspondence and reprint requests to Dr. Jeremiah Tilles, Department of Medicine, University of California, Irvine, Building 53, Route 81, 101 South City Drive, Orange, CA 92868, U.S.A. Manuscript received August 7, 1996; accepted May 12, 1997. suggested that antiretroyiral therapy can improve clinical outcomes such as mortality, occurrence of opportunistic infections, and progression to AIDS (6,7). Combination therapy seems to be superior to monotherapy (8). Unfortunately, current therapeutic regimens result in suppression but not eradication of HIV. Furthermore, antiretroviral therapy is limited by both drug toxicity and the invariable development of resistance. A variety of studies have demonstrated that resistance is associated with specific mutations in the HIV pol gene (9-12). If it were possible to directly monitor the occurrence of such mutations in the HIV RNA sequence of a given patient, specific alternative therapies might be considered before a surrogate marker (e.g., CD4 count or viral load) could be expected to even reflect a failure of the current regi- men. However, detection of drug resistance may only be indicated for those patients with a rise in viral load on therapy based on studies in which continued suppression of HIV RNA was associated with a wild-type genome 356 (7). Furthermore, the clinical application of sequence in- formation in an individual patient may not currently be feasible because of the time and labor required for the sequencing process and subsequent data management. Fortunately, computer technology can be applied to solve the problem of bulky sequence data management. Rule-based expert systems (13,14) are computer pro- grams that declaratively represent knowledge of a spe- cialized problem and facts about a specific case? and draw inferences about the case. In the program devel- oped for the current study (Customized Treatment Strat- egies for HIV [CTSHIV]), the rules encode information on drug-resistant mutations of HIV and characteristics of current antiviral agents, the facts are the sequences of the HIV genomes obtained from a specific individual, and the inference to be drawn is a set of recommended drug regimens. The knowledge of a specific problem may be represented as a set of weighted if-then rules of the form: IF THEN . For example, one such rule in CTSHIV is: IF Leucine is encoded by codon 4 1 of the RT portion of the pol gene, then do not use AZT (weight = 0.3). The rule is derived from the literature, which reports that a methionine-to- leucine mutation at codon 41 confers approximately fourfold resistance to AZT (15). The rules are weighted from 0.1 (low priority) to 1.0 (high priority) based on level of resistance or support in the literature. To draw an inference about a particular conclusion (e.g., which drugs should not be used), the expert system finds rules whose consequence matches that conclusion, and attempts to determine whether the antecedent of the rule is true. If the antecedent can be shown to be true, then the conclu- sion is asserted to be true. There are two ways in which an antecedent may be true. First, the antecedent may match a fact that is known. For example, in the rule just stated, this would check to see whether there is a leucine encoded by codon 41 of the RT portion of the pol gene. Second, the antecedent may match the consequence of another rule, and the antecedent of that rule can then be shown to be true. The current study reports a process by which the pro- tease, reverse transcriptase, and integrase segments of the HIV pol gene can be cloned using three primer sets and the sequences downloaded into an expert system that recommends multiple ranked two-, three-, and four-drug regimens based primarily on the occurrence of known mutations coding for specific drug resistance. METHODS Cloning and Sequencing Blood was obtained from 10 HIV-infected individuaIs after informed consent was obtained using a form approved by the Institutional Re- view Board at UC Irvine, All patients had CD4 counts <500/mm3, two had never received antiretroviral therapy, and the others had received a variety of regimens, Standard reverse transcription-polymerase chain reaction (RT-PCR) was performed using published primer sets for pro- tease (16) and RT (17) and primers developed in our laboratory for integrase (upstream 5’CAAGTAGATAAATTAG TCAGTGCTG- GAATCAGG-3’: downstream 5’.CCTAGCTTTCCCTGAAACATA- CATATGGTG’IT3’). PCR was performed on all 10 patients to dem- onstrate reproducibility of the primer sets used. Because cloning and sequencing of PCR products is well standard- ized, PCR product from only two patients was sequenced. Briefly. a portion of PCR product (calculated to give an approximate vector insert ration of 1: 1) was Iigated into a T/A vector (Invitrogen). Two micro- liters of the ligation reaction was transferred to a tube containing 50 p,l of competent Eschen’chia coli cells. The bacteria were transformed (18), plated onto Luria-Bertani (LB) agar plates containing 50 pg/ml of kanamycin. and grown at 37°C for 72 h. For each plate, a single white colony was grown overnight in 5 ml of liquid LB broth containing 50 kg/ml of kanamycin. The plasmid DNA was purified and extracted using a Qiagen plasmid preparation kit (Chatsworth, CA). For the two patients, five clones per PCR product were sequenced using a dye- termination process (19) with an automated ABI sequencer. One was asymptomatic and had a CD4 count of 285/mm3 without prior antiret- roviral therapy. The other patient had a history of Pneumucvstis carinii pneumonia and mild wasting syndrome, had received multiple antiret- roviral regimens, and currently had a CD4 count of 55 cells/mm’. Development of Rules for the Expert Program Base substitutions coding for either in vitro and/or resistance to specific drugs (20-24) were identified in the literature and included as rules for the expert system. The rules were weighted based on the degree of resistance. Additional rules ranked available therapies based on HIV antiviral activity, overlapping toxicities, and mechanisms of action. Currently, there are 48 rules in the program (Table 1). Development of the Expert Program The CTSHIV system is implemented in first-order computer learning (FOCL-1-2-3) (14), a backward chaining expert system for the Mac- intosh computer. The facts needed to determine which drugs to exclude are downloaded in CTSHIV. The recommendation process goes through six distinct phases: 1. Exclusion of antiretroviral agents based on sequence information. The medical literature contains an increasing number of studies on the relationship between mutations and drug resistance. 2. Inclusion of drugs with enhanced antiretroviral activity associated with specific genetic sequences (25). 3. Exclusion of antiviral combinations with overlapping toxicities, redundant mechanisms of action, or unproven efficacy and/or safety (e.g., NNRTIs used in combination with protease inhibi- tors). 4. Weighting the aforementioned combinations from 0.1 (low pri- ority) to 1.0 (high priority) based on either: a. level of resistance conferred (phase 1; 2-fold or >lOO-fold resistance were assigned a value of 0.1 or 1.0, respectively; intermediate resistance was assigned a value between 0.1 and 1.0 [22,23]), or b. strength of supporting evidence in the literature (phases 2-3 125,261). EXPERT SYSTEM IN MANAGEMENT OF HIV-INFECTED PATIENTS 357 JournaI of Acquired Immune Dejkienc)- Syndromes and Human Retroviroiog~, Vol. IS, No. 5, 1997 3.58 M. J. PAZZANI ET AL. TABLE 1. Complete set qf current rules qf the Customized Treatment Strategies for HIV expert program 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. AZT codon 41 RT (CTT CTC TTA CTA CTG TTG) [0.3] (15) AZT codon 41 RT (CM CTC TTA CTA CTG TTG) and codon 215 RT (TTC TAC) [l.O] (31) AZT codon 67 RT (AAT AAC) and codon 70 RT (AGA AGG CGG CGC) and codon 215 RT (TTC TAC) and codon 219 RT (CAA CAG) [0.8] (32) AZT codon 41 RT (CTT CTC TTA CTA CTG TTG) and codon 67 RT (AAT AAC) and codon 70 RT (AGA AGG CGG CGC) and codon 215 RT (TTC TAC) [I.01 (32) AZT codon 67 RT (AAT AAC) IO.61 (33) AZT codon 70 RT (AGA AGG CGG CCC) [0.6] (33) AZT codon 215 RT (TTC TTT) [0.7] (33) AZT codon 219 RT (CAA CAG GAA GAG) [O.l] (33) ddl codon 74 RT (GTA GTG GTT) [0.4] (20) (ddl ddC) codon 184 RT (GTA GTG GTT) [0.2] (34) (AZT ddl d4T ddC) codon 151 RT (ATG) [l.O] (35) AZT codon 210 RT (TGC) and codon 215 (TTC ‘ITT) [0.9] (36) (AZT ddl d4T ddC) codon 69 RT (GGA GGT GGG) [0.6] (37) 3TC codon 184 RT (GTA GTG GTT) [l.O] (34) 3TC codon 184 RT (ATA ATC ATT) [0.6] (34) (Nevirapine Delavirdine) codon 103 RT (AAT AAC) [0.5] (38) Nevirapine codon 106 RT (GCA GCT GCC) [0.9] (39) Nevirapine codon 108 RT (ATA ATC ATT) [0.5] (40) (Nevirapine Delavirdine) codon 181 RT (TGT TGC) [0.9] (41) Nevirapine codon 181 RT (ATA ATC ATT) LO.71 (42) Nevirapine codon 188 RT (TGT TGC) [0.5] (40) Nevirapine codon 190 RT (GCA GCT CCC) [OS] (39) Delavirdine codon 236 RT (CM CTC TTA CTA CTG TTG) W.61 (43) Delavirdine codon 69 RT (GGA GGT GGG) [0.2] (37) Nevirauine codon 69 RT (CGA GGT GGG) 10.41 (37) Ritonaiir codon 8 Pro (CAA CAG AAA AkG) [0.4]‘(44) Ritonavir codon 32 Pro (ATA ATC ATT) [0.3] (45) (Ritonavir Saquinavir) codon 48 Pro (GTA GTG GTT) [0.6] (46) (Ritonavir Indinavir) codon 82 Pro (GCA GCT GCC TTC TTT) LO.61 (47) 30. Ritonavir codon 10 Pro (ATA ATC ATT) [0.4] (47) 31. Ritonavir codon 54 Pro (GTA GTG GTT) [0.4] (47) 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46 47 48. Indinavir codon 46 Pro (ATA ATC ATT CTT CTC TTA CTA CTG TTG) [0.6] (48) Indinavir codon 82 Pro (ACA ACT ACC) [0.6] (48) (Ritonavir Indinavir Saquinavir) codon 84 Pro (GTA GTG GTT) 10.61 (47) Indinavir codon 32 Pro (ATA ATC ATT) [0.2] (48) (Indinavir Ritonavir Nelfinavir) codon 71 Pro (GTA GTG GTT) [O.l] (47,49) Saquinavir codon 90 Pro (ATG) [0.6] (50) Saquinavir codon 48 Pro (GTA GTG GTT) and codon 90 Pro (ATG) [l.O] (50) Neltinavir codon 30 Pro (AAT AAC) [0.8] (49) USE Nevirapine if codon 236 RT (CTT CTC TTA CTA CTG TTG) [0.5] (43) Do not use Nelfinavir with Indinavir, Ritonavir, or Saquinavir t1.01 (51) Do not use Indinavir with Ritonavir or Saquinavir [l .O] (51) Do not use both Delavirdine and Nevirapine together [0.9] (8) Do not use both ddc and ddl together [ 1.01 (51) Do not use either Nelfinavir, Indinavir, Ritonavir, or Saquinavir with either Nevirapine or Delavirdine [l.O] (51) The expert program will receive five sequences from each patient. If all five cause a rule to fire, then multiply the weight of rule by 1.0; if four of five cause the rule to fire, multiply the weight by 0.8; if three of five cause the rule to fire, multiply the weight by 0.6; if two of five cause the rule to fire, multiply the weight by 0.4; and if one of five cause the rule to fire, multiply the weight by 0.2. Display five two-drug regimens, five three-drug regimens. and five four-drug regimens. If more regimens are permissible, delete the following drugs in order of preference: ddC, ddl Nevirapine, d4T, Saquinavir, Delavirdine, AZT, 3TC, and Nelfinavir. When more than one two, three, or four drug regimens are displayed, rank each regimen by providing a value for each drug within the regimen based on the sum of the weights accumulated from the rules. Rank in descending order the regimens with the lowest total value. In case of equal value: rank in the following order: Indinavir, Ritonavir. Nelfinavir. 3TC, AZT, Delavirdine, Saquinavir, d4T, Nevirapine, ddl. ddC. The rules were developed from information in the medical literature and integrated into the FOCL expert system shell. Most of the rules are exclusionary, based on drug-resistant mutations. For example, rule 1 is interpreted as, “Do not use AZT if the nucleotides at codon 41 of the reverse transcriptase portion of the polymerase gene are ‘CCT CTC TTA CTA CTG or TTG.’ ” The parentheses containing the names of drugs or codons represent an “or” between the elements within them. Weights are in brackets, scaled from 0.1 (lowest priority) to 1.0 (highest priority), Exclusionary rules are weighted by the degree of drug resistance conferred. Drug interaction rules are weighted by the severity of overlapping toxicities. Disallowed drug combinations are weighted by the relative lack of supporting evidence in the literature for their use. Each weighted rule is referenced. AZT, azidothymidine: ddl, didanosine; ddC, dideoxycytidine: d4T, starvudine; 3TC, lamivudine. 5. Adjusting the weight of each conclusion by the proportion of clones causing the rule to fire. 6. Identifying and ranking combinations of candidate drugs based on antiviral activity. RESULTS Recommended regimens for the study patients: HIV nucleic acid was successfully amplified from all 10 pa- tients using each of the three primer pairs. Complete sequences of five clones from two patients were down- loaded into the CTSHIV program and selected regimens choice @a&e 2). Ali five clones frdm patient 1 (CD4 were disulaved in coniunction with exDlanations for their count of 285/mm’, no prior antiretrovirals) had identical sequences, including 14 and 6 amino acid substitutions for RT and Pro, respectively, compared with HXB2 (Fig. 1). None of the changes resulted in the firing of a rule. In contrast, there was great heterogeneity among the five clones from patient 2 (CD4 count of 55/mm3, multiple antiretroviral regimens in the past). There was an average of 47 and 22 amino acid changes for RT and Pro, re- spectively, firing eight exclusionary rules in the expert program. EXPERT SYSTEM IN MANAGEMENT OF HIV-INFECTED PATIENTS TABLE 2. Protocols recommended for patients I and 2 Protocols recommended for patient 1 The following protocols with tu‘o drugs are recommended: Indinavir 3TC Indinavir AZT Ritonavir 3TC Ritonavir AZT 3TC AZT The following protocols with three drugs are recommended: Indinavir 3TC AZT Ritonavir 3TC AZ7 Ritonavir 3TC Saquinavir Ritonavir AZT Saquinavir 3TC AZT Delavirdine The following protocols with four drugs are recommended: Indinavir 3TC AZT D4T Rotonavir 3TC AZT Saquinavir Ritonavir 3TC AZT D4T Ritonavir 3TC Saquinavir D4T Ritonavir AZT Saquinavir D4T Protocols recommended for patient 2 The following protocols with two drugs are recommended: 3TC Delavirdine 3TC d4T 3TC Nevirapine Delavirdine d4T 3TC ddC The following protocols with three drugs are recommended: 3TC Delavirdine d4T 3TC d4T N&rapine 3TC Delavirdine ddC 3TC d4T ddC 3TC N&rapine ddC The following protocols with four drugs are recommended: 3TC Delavirdine d4T ddC 3TC d4T Nevirapine ddC 3TC Delavirdine d4T ddl 3TC d4T Nevirapine ddl 3TC d4T ddC Indinavir 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.4 0.6 By rule I, AZT reduced by 0.12 because two sequences had TTG at codon 41 of the RT gene. Kellam P, Boucher C, Larder BA. Fifth mutation in human immunodeficiency virus type 1 reverse transcriptase contributes to the development of high-level resistance to zidovudine. Proc Nail Acad Sci U.SA 1992:89: 1934-8. By rule 28, Ritonavir reduced by 0.24 because two sequences had GTG at 0.6 codon 48 of the Pro gene. Boucher C. Rational approaches to resistance: Using 5aquinavir. AIDS 1996;1O(suppl l):S15-9. By rule 5, AZT reduced by 0.6 because five sequences had AAC at codon 67 of the RT gene. Cal&do A, Savara A, An D, DeVore K, Kaplan J, D’Aquila R. Effects of zidovudine-selected human immunodeficiency virus type 1 reverse transcriptase amino acid substitutions on processive DNA synthesis and viral replication. J Vi& 1996:70:2146-53. By rule 34, Indinakir, Saquinavir, Ritonavir reduced by 0.6 because five sequences had GTG at codon 84 of the Pro gene. Schmidt J, Ruiz L, Clotet B, et al. Resistance-related mutations in the HIV-1 protease gene of patients treated for I year with the protease inhibitor ritonavir (ABT-538). AIDS 1996:10:995-9. By rule 38, Saquinavir reduced by 0.4 because two sequences had GTG at codon 48 of the Pro gene and ATG at codon 90 of the Pro gene. Jacobsen H, Haenggi M, Ott M, Duncan I, Andreoni M, Vella S, Mous 1. Reduced sensitivity TO saquinavir: An update on genotyping from phase III1 trials. Anrivirai Res 1996;29:95-7. By rule 9, ddl reduced by 0.4 because five sequences had GTA at codon 74 of the RT gene. St Clair M, Martin J, Tudor W. Resistance to ddl and sensitivity to AZT induced by a mutation in HIV-1 reverse transcriptase. Science 1991;253:1557-9. By rule 6, AZT reduced by 0.6 because five sequences had AGA at codon 70 of the RT gene. Caliendo A, Savara A, An D, DeVore K, Kaplan J, D’Aquila R. Effects of zidovudine-selected human immunodeficiency virus type 1 reverse transcriptase amino acid substitutions on processive DNA synthesis and viral replication. J Viral 1996;70:2146-53. By rule 37. Saquinavir reduced by 0.36 because three sequences had ATG at codon 90 of the Pro gene. Jacobsen H, Haenggi M, Ott M, Duncan I, Andreom M, Vella S, Mous J. Reduced wnsitivity to saquinavir: An update on genotyping from phase I/I1 trials. Antiviral Rer 1996;29:95%7. DISCUSSION Antiretroviral therapy is useful in slowing down the progression of HIV infection. However, all therapeutic regimens studied to date eventually fail, resulting in clinical deterioration of the patient and worsening of sur- rogate markers such as a rise in CD4 count, an increase in viral load, or both. The current standard of practice indicates a change in therapy when this occurs. However, guidelines as to specific changes in antiretrovirals is lacking, and the possibility of appropriately changing therapy for specific mutants before surrogate marker de- terioration is possible. The time to onset of clinically important mutations is highly variable. Cross-resistance occurs between agents in the same class or with similar mechanisms of action (21). Finally, drug-specific resis- tance mutations also occur in wild-type strains (22). Thus, basing changes in therapy on drug sensitivity would seem to have merit. The results of the current study demonstrate the feasibility of such an approach using an expert program to recommend treatment regi- mens. The program also encodes information about level of drug resistance, overlapping toxicities, and other fac- tors. Learning algorithms such as FOCL (14) can be applied to the program to modify it based on outcome measures from patients such as CD4 count, viral load, and progression to AIDS. Old rules can be reweighted or eliminated. New rules can be added based on information from the literature or the development of new antiretro- viral agents. The HIV strains in an individual patient are not ge- nomically homogenous. The current system allows de- tection of a resistant mutation if present in at least one of five clones. If detection of crucial mutations in a patient found less frequently than 20% is later thought to be important, hybridization techniques (27) could be ap- plied to capture those present in as few as 1 of 20 strains. The CTSHIV system uses primers from conserved re- gions, yielding detectable PCR products for each of the 10 patients tested. Furthermore, it contains an expert sys- tem to effectively manage the sequence data and recom- mend specific therapeutic regimens based on genetic se- quences and information from the medical literature. Nucleotide sequences could be aligned using software programs (28), reducing the chance of downloading in- determinate foci. However, the current system incorpo- Journal of Acquired Immune Deficiency Syndromes and Human Retrovimlogy, Vol. 15, No. 5, 1997 360 M. J. PAZZANI ET AL. position 1 11 21 31 41 51 61 71 HXB2 CCTCAGGTCA CTCTTTGGCA ACGACCCCTC GTCACAATAA AGATAGGGGG GCAACTAAAG GAAGCTCTAT TAGATACAGG Patient 1 clone1 ---- __A___ -___ -___ __ _____ __._. -- _G______ ___.______ __________ __________ __________ clone2 ______A___ ____ ____._ ____ ______ -_ _G______ ___.______ __________ __________ __________ clone3 ---- __A___ __________ _______.__ ___G___-__ _.________ ____.__-__ __________ -_________ clone4 ---- __A___ ____.___-_ __________ ---G--_--u _._.___.__ __________ __________ _______-__ clone6 ______A.__ ____ ______ _______.__ __ _G______ __________ __________ __________ __________ Patient 2 clone1 ---- __A___ ____C_____ _______G__ ___T______ G-C_______ ______G___ ____________________ clone2 ---- __A___ ____ C _____ _______G__ ___T______ G_C____A__ ______G___ _________________-__ clone3 ---- __A___ ____C_____ __________ ___T___.__ __C____A__ ____._G___ ____________________ clone4 ---- __A___ ____C_____ _______G__ ___T______ G_C____A__ ______G___ ____________________ clone5 ---- __A___ _-__C__--_ __________ ___T.__.__ __C____A__ ____._G___ ____________________ 81 91 101 111 121 131 141 151 HXB2 AGCAGATGAT ACAGTATTAG AAGAAATGAG TTTGCCAGGA AGATGGAAAC CAAAAATGAT AGGGGGAATT GGAGGTlTTA Patient 1 clone, __________ ------____ __________ -__-___.__ _A________ -._-_-_--. ___.______ ___--__--- clone2 ___.______ -----_---. ___.__.__. ---------_ _A________ __________ __________ __________ clone3 ___.______ ___.______ ___.______ -__-___-._ _A__________________ __________ __________ cbne4 __._______ _____.____ ___._.____ _______.__ _A________ _______-__ __________ -__-__---_ clone5 __________ ____-_____ __________ ___-__..__ _A________ ___-___.__ __________ __________ Patient 2 clone, _ _ _ _ _ _ _ _ _ _ - _ - _ - - - - - - T_______G_ __________ __________ __________ __________ --.--.---- c,,3ne2 __________ .___._____ T_______G_ __________ __________ __________ __________ __________ clone3 __________ __________ _.________ ____________________ __________ __T_______ __________ clone4 ___ .._____ ____-__--_ T_______G_ __________ __________ ______.___ __________ __________ clone5 __..______ __________ __________ ______.___ __________ __________ __T_______ _.________ 161 171 161 191 201 211 221 231 HXB2 TCAAAGTAAG ACAGTATGAT CAGATACTCA TAGAAATCTG TGGACATAAA GCTATAGGTA CAGTATTAGT AGGACCTACA Patient 1 clone1 _________.__________ ____-_____ ____.._-_. ---- __C___ ____,.__I^ __A__ ____ _“_I__.___ clone2 ____________________ __________ ____-__-______ __C___ _________. __A_______ __________ clone3 ____________________ ___--..-.- --------__ .--_ __C___ __________ __A-______ __________ clone4 _________.__________ __________ -______-__ .--- __C___ ___________.A_______ __________ clone6 ___ ._._ __- __________ ___-______ ____--_-__ ---- __C___ ____._____ __A_______ ___.______ Patient 2 clone, __________CA________ __________ ___ __T___A__________ -A___TC__C__________ __________ ,3,3-,e2 __________CA________ _-_____-_- _C___T___,+__________ _A__.TC__C __________ __________ clone3 __________CA________ -___-__-_- _-__ _T____ __________ _A____C___ _______.__ __________ clone4 __________ CA________ __________ _____T___A __________ _,t,___TC_-C __________ __________ c,on*5 __________ CA________ __________ _____T____ __________ _A____C___ __________ __________ 241 251 261 271 261 291 HXBP CCTGTCAACA TAATTGGAAG AAATCTGTTG ACTCAGATTG GTTGCACTTT AAATTTT Patient 1 c,o,,e, __________ _.____________.__.__ __________ _G________ _______ clone2 __________ ____________________ __________ _G________ _______ clone3 __________ ____________________ __________ _G________ _______ clone4 __________ ______________-_____ ._________ _G________ _______ clone5 __________ __________-___-_____ __________ _G________ _______ Patient 2 clone, _________ =_________________A__ ________G_ ________C_ _______ clone2 _____ ___. G____________________ _______.G_ ________C_ _______ .-lone3 _________G____________________ ________G_ ________C_ ____.__ clone4 _________o_______ ______ _-- _A__ ________G_ ________C_ _______ clone5 _._ ______ 0 ._.._____________A__ ________G_ ________C_ _______ FIG. 1. Complete sequence of the protease portion of the human immunodeficiency virus polymerase gene (297 bp) for each of five clones from two patients. Patient 1 has never received antiretroviral therapy, and has a current CD4 count of 285/mm3. Patient 2 has received multiple combinations of azidothymidine, didanosine, lamivudine, and Saquinavir. The sequences are compared with that of strain HXBP (Los Alamos data bank). The sequences that cause a rule to fire are in bold type. J~~mal of Acquired Immune Deficiency Syndromes and Human Retrovirology, Vol. 15, No. 5, 19%’ EXPERT SYSTEM IN MANAGEMENT OF HIV-INFECTED PATIENTS 361 rates sequencing of multiple clones. Although our pre- liminary studies suggested that such a process both re- duces the display of indeterminate residues and enhances the opportunity to identify drug-resistant strains repre- senting a small proportion of the total number of quasi- species in an individual patient compared with sequenc- ing the heterogenous mixture of strains, it also adds sig- nificantly to the time and cost. In this pilot study, every effort was made to yield data as cleanly as possible. However, if the system were to be used in the everyday management of patients, sequencing of uncloned, ampli- fied products would be more practical and may yield comparable results. Specialty Laboratories (Santa Monica, CA, U.S.A.) is currently offering genotyping of heterogenous PCR product commercially. Alternatively, rapid sequencing of the polymerase and protease genes in development by Affymetrics (Sunnyvale, CA, U.S.A.) and Applied Biosystems (Foster City, CA, U.S.A.) using fluorescent primer-based cycle sequencing (29) may ren- der the use of the CTSHIV expert system more feasible. The CTSHIV system is limited by an incomplete un- derstanding of the correlation between genomic muta- tions conferring drug resistance and clinical and surro- gate marker outcomes. Furthermore, the expert program can only incorporate rules from the current literature and recommend licensed drugs. A study to correlate surro- gate marker response and clinical outcomes with the use of the program is in progress. In addition, a study by the CPCRA is under way to assess the clinical utility of genotyping (Volberding P, personal communication, February 1997). Studies with large numbers of patients will ultimately be required to validate the CTSHIV sys- tem and determine whether it may assist in the everyday management of HIV-infected individuals and in the de- sign, implementation, and interpretation of outcomes in clinical trials (30). Currently, however, there is no evi- dence that the system will perform better than clinical judgment. Acknowledgments: The technical assistance of Tonya Clark is appreciated. The authors thank Douglas Richman, M.D. for reviewing and revising the expert rules. The sequencing was performed by Erik Avajani and Clifford Brunk. This work was supported by the California Universitywide AIDS Research Program through the California Collaborative Treatment Group (CCTG). REFERENCES 1. DeClercq E. Antiviral therapy for human immunodeficiency virus infections. Clin Microbial Rev 1995;8:200-39. 2. Frost S, McLean A. Quasispecies dynamics and the emergence of drug resistance during zidovudine therapy of HIV infection. AIDS 1994;8:323-32. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13 14 15. 16. 17. 18. 19. 20. 21. 22. Katlama C, Ingrand D, Loveday C, et al. Safety and efficacy of lamivudine-zidovudine combination therapy in antiretroviral-naive patients. MMA 1996;276: 118-25. Holodniy M, Katzenstein D, Winters M, et al. Measurement of HIV virus load and genotypic resistance by gene amplification in asymptomatic subjects treated with combination therapy. / Acquir Immune Defic Syndr 1993;6:366-9. Kahn J, Lagakos S, Richman D, et al. A controlled trial comparing zidovudine with didanosine in human immunodeficiency virus in- fection: The NIAID AIDS Clinical Trials Group. N Engl J Med 1992;327:581-7. Schooley RT. Ramirez-Ronda C, Lange JM, et al. Virologic and immunologic benefits of initial combination therapy with zidovu- dine and zalcitabine or didanosine compared with zidovudine monotherapy: Wellcome Resistance Study Collaborative Group. J infect Dis 1996;173:1354-66. Zhu QY, Scarborough A, Polsky B: Chou TC. Drug combinations and effect parameters of zidovudine, stavudine, and nevirapine in standardized drug-sensitive and resistant HIV type 1 strains. AIDS Res Hum Retroviruses 1996;12:507-17. Lange JM. Triple combinations: Present and future. J Acquir Zm- mw-ze Defic Syndr Hum Retrovirol 1995;1O(suppl l):S77-82. Nielsen C, Bruun L, Mathiesen L. Pedersen C, Gerstoft J. Devel- opment of resist;mce to zidovudine (ZDV) and didanosine (ddl) in HIV from patients in ZDV, ddl and alternating ZDV/ddl therapy. AIDS 1996;10:625-33. Larder B. Kellam P, Kemp S. Zidovudine resistance predicted by direct detection of mutations in DNA from HIV-infected lympho- cytes. AIDS 1991;5:13744. Larder B, Kemp S. Multiple mutations in HIV-l reverse transcrip- tase confer high-level resistance to zidovudine (AZT). Science 1989;246:1155-8. Frenkel L. Wagner M, Atwood S, Cummins T, Dewhurst S. Spe- cific, sensitive, and rapid assay for HIV pol mutations associated with resistance to zidovudine and didanosine. J Infect Dis 1995; 33:342-7. Jackson P. Introduction to expert systems. Reading, MA: Addison- Wesley, 1990. Pazzani M, Kibler D. The utility of knowledge in inductive leam- ing. Machine Leatnlizg 1992;9:57-94. Kellam P, Boucher C, Larder BA. Fifth mutation in human immu- nodeficiency virus type 1 reverse transcriptase contributes to the development of high-level resistance to zidovudine. Proc Nat1 Acad Sci USA 1992;89:1934-8. Fontenant G, Johnston K, Cohen J, Gallaher W, Robinson J, Luftig R. PCR amplification of HIV-l proteinase sequences directly from lab isolates allows determination of five conserved domains. Viral 1992;190:1-10. Gu Z, Gao Q, Fang H, et al. Identification of a mutation at codon 65 in the IKKK motif of reverse transcriptase that encodes human immunodeticiency virus resistance to 2’,3’-dideoxycytidine and 2’,3’-dideoxy-3’.thiacytidine. Antimicrob Agents Chemother 1994;38:275-81. Clark J. Novel non-templated nucleotide addition reactions cata- lyzed by procaryotic and eucaryotic DNA polymerases. Nucleic Acid Res 1988;20:9677-86. Giesecke H, Obermaier B, Domdey H. Rapid sequencing of the Sendai virus 6.8 kb large (L) gene through primer walking with an automated DNA sequencer. J Viral Methods 1992;38:47-60. St Clair M, Martin J, Tudor W. Resistance to ddl and sensitivity to AZT induced by a mutation in HIV- 1 reverse transcriptase. Science 1991;253:1557-9. Gu Z, Gao Q, Li X, Parniak M, Wainberg M. Novel mutdon in the human immunodeficiency virus type 1 reverse transcriptase gene that encodes cross-resistance to 2’,3’-dideoxyinosine and 2’,3’- dideoxycytidine. J Viral 1992;66:7128-35. Gu Z, Gao Q, Fang H, Pamiak M, Brenner B, Wainberg M. Iden- tification of novel mutations that confer drug resistance in the Jounud ofAcquired hmune Deficiency Syndromes and Human Retrovirology, Vol. 15, IVO. 5, 1997 M. J. PAZZANI ET AL. human immunodeficiency virus polymerase gene. Leukemia 1994; 8S1:5166-9. 23. Richman D, Shih C, Lowy 1: Rose J, Prodanovich P, Goff S, Griffin J. Human immunodeficiency virus type 1 mutants resistant to nonnucleoside inhibitors of reverse transcriptase arise in tissue culture. Proc Nat1 Acad Sci USA 1991;88:11241-5. 24. El-Farrash M, Kuroda M, Kitazaki T, Masuda T. Kato K, Hatanaka M, Harada S. Generation and characterization of a human immu- nodeficiency virus type 1 (HIV-l) mutant resistant to an HIV-l protease inhibitor. J Viro[ 1994;68:23339. 25. Dueweke T, Pushkarskaya T, Poppe S, et al. A mutation in reverse transcriptase of bis(heteroaryl)piperazine-resistant human immu- nodeficiency virus type 1 that confers increased sensitivity to other non-nucleoside inhibitors. Proc Nutl Acad Sci USA 1993;90: 4713-7. 26. Hammer S, Kessler H, Saag M. Issues in combination antiretrovi- ral therapy: a review. J Acquir Immune Defic Syndr 1994;7 Suppl 2:S2435. 27. Holodniy M, Mole L, Margolis D, et al. Determination of human immunodeficiency virus RNA in plasma and cellular viral DNA genotypic zidovudine resistance and viral load during zidovudine- didanosine combination therapy. J Viral 1995;69:3510-6. 28. Michael N, Davis KE, Loomis-Price L, VanCott T, Burke D, Red- field R, Birx D. V3 seroreactivity and sequence variation: Tracking the emergence of V3 genotypic variation in HIV-l-infected pa- tients. AIDS 1996;10:121L9. 29. Luehrsen KR, Marr L, van der Knaap E, Cumberledge S. Analysis of differential display RT-PCR products using fluorescent primers and GENESCAN software. Biotechaiques 1997;22:168-74. 30. Ohno-Machado L, Parra E, Henry S, Tu S, Musen M. AIDS 2: A decision-support tool for decreasing physician’s uncertainty re- garding patient eligibility for HIV treatment protocols. Proceed- ings of the 17th Annual Symposium on Computer Applications in Medical Care, Washington, DC, 1993:429-33. 3 1. Boucher C, O’Sullivan E, Mulder A, et al. Ordered appearance of zidovudine resistance mutations during treatment of 18 human immunodeflciency virus-positive patients. J Infect Dis 1992;165: 105-10. 32. Gurusinghe A, Land S, Birch C, et al. Reverse transcriptase mu- tations in sequential HIV-l isolates in a patient with AIDS. J Med Virol 1995;46:23843. 33. Caliendo A: Savara A, An D, DeVore K, Kaplan J, D’Aquila R. Effects of zidovudine-selected human immunodeficiency virus type 1 reverse transcriptase amino acid substitutions on processive DNA synthesis and viral replication. J Viral 1996;70:2146-53. 34. Quan Y, Gu Z, Li X, Li Z, Morrow C, Wainberg M. Endogenous reverse transcription assays reveal high-level resistance to the tri- phosphate of (-)2’-dideoxy-3’-thiacytidine by mutated M184V hu- man immunodeficiency virus type 1. J Viral 1996;70:5642-5. 35. Iversen AK, Shafer RW, Wearly K, Winters MA, Mullins JI, Che- sebro B, Morgan TC. Multidrug-resistant human immunodeficien- cy virus type 1 strains resulting from combination antiretroviral therapy. J Viral 1996;70: 1086-90. 36. Hooker D, Tachedjian G, Soloman A, et al. An in viva mutation from leucine to tryptophan at position 210 in human immunode- 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. ficiency virus type 1 reverse tmnscriptase contributes to high-level resistance to 3’.azido-3’-deoxythymidine. J Vi& 1996;70:8010-8. Kew Y, Saloman H, Olsen L, Wainberg M, Prasad V. The nucleo- side analog-resistant E89G mutant of human immunodeficiency virus type 1 reverse transcriptse displays a broader cross-resistance that extends to nonnucleoside inhibitors. Anfimicrob Agents Che- mother 1996;40:17114. Zhu Q, Scarborough A, Polsky B, Chou T. Drug combinations and effect parameters of zidovudine, stavudine, and nevirapine in stan- dardized drug-sensitive and resistant HIV type 1 strains. AIDS Res Hum Retroviruses 1996;12:507-17. Richman D, Havlir D, Corbeil J, et al. Nevirapine resistance mu- tations of human immunodeficiency virus type 1 selected during therapy. / Viral 1994;68:1660-6. Richman DD. Resistance of clinical isolates of human immunode- ficiency virus to antiretroviral agents. Antimicrob Agents Che- mother 1993;37:1207713. Richman D. Shih CK, Lowy I, Rose J. Prodanovich P, Goff S, Griffin J. Human immunodeflciency virus type 1 mutants resistant to nonnucleoside inhibitors of reverse transcriptase arise in tissue culture. Proc Nat1 Acad Sci USA 1991;88:11241-5. Havlir, Eastman S, Gamst A, Richman D. Nevirapine-resistant human immnodeficiency virus: Kinetics of replication and esti- mated prevalence in untreated patients. J Viral 1996;70:7894-9. Dueweke TJ, Pushkarskaya T, Poppe SM. et al. A mutation in reverse transcriptase of bis(heteroary1) piperazine-resistant human immunodeticiency virus type 1 that confers increased sensitivity to other non-nucleoside inhibitors. Proc Nat1 Acad Sci USA 1993;90: 4713-7. Ho DD, Toyoshima T, MO H, et al. Characterization of human immunodeficiency virus type 1 variants with increased resistance to a CZ-symmetic protease inhibitor. J Viral 1994;68:201&2020. Kaplan AH, Michael SF, Wehbie RS, et al. Selection of multiple human immunodeficiency virus type 1 variants that encode viral proteases with decreased sensitivity to an inhibitor of viral prote- ase. Proc Nat1 Acad Sci USA 1994;91:5597-601. Boucher C. Rational approaches to resistance using saquinavir. AIDS 1996;1O(suppl l):S15-9. Schmidt J, Ruiz L, Clotet B, et al. Resistance-related mutations in the HIV-l protease gene of patients treated for 1 year with the protease inhibitor ritonavir (ABT-538). AIDS 1996;10:995-9. Condra _I, Holder D, Schlief W, et al. Genetic correlates of in vivo viral resistance to indinavir, a human immunodeficiency virus type 1 protease inhibitor. J Viral 1996;70:8270-6. Patick A, Duran M, Cao Y, et al. Genotypic analysis of HIV-l variants isolated from patients treated with the protease inhibitor Neltinavir, alone or in combination with d4T or AZT and 3TC [Abstract]. Presented at the Fourth Conference on Retroviruses and Opportunistic Infections, Washington, DC, 1997. Jacobsen H, Haenggi M, Ott M, Duncan I, Andreoni M, Vella S, Mous J. Reduced sensitivity to saquinavir: An update on genotyp- ing from phase I/II trials. Antiviral Res 1996;29:9557. Taburet A, Singlas E. Drug interactions with antiviral drugs. Clin Pharmacokinet 1996;30:385-401. J*umal of Acquired Immune Deficiency Syndromes and Human Retrovirology, Vol. 15, No. 5, 1997