INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Human immunodeficiency virus type 1 (HIV-1) infection risk factors are usually identified directly from a patient's medical history. However, additional evidence is essential when viral transmission occurs during clinical care (8, 16, 25, 37) or is the object of a criminal trial (1, 48). Because nucleotide sequences vary greatly among HIV-1 isolates (28, 41, 49), transmission histories among infected individuals can be investigated through estimated phylogenetic relationships among viral sequences.

Previous studies have shown that phylogenetic methods can lead to accurate reconstruction of known transmission histories (5, 11, 33, 52). Phylogeny inferences have also been used to reconstruct unknown transmission histories (1, 7, 8, 37), although the methodology itself has sometimes been a point of contention. For example, in the Florida dentist case (37), while the original analysis determined HIV transmission, reanalyses of the same data set together with new control sequences did not unequivocally reject the possibility of independent acquisitions of similar local variants (6, 14). However, extensive analyses (using multiple inference methodologies) of an extended data set convincingly confirmed the existence of a dental clade (13, 24). Furthermore, phylogenetic analyses are increasingly used in court (2, 48) and must therefore "rise to a threshold of reliability" (48).

Investigating a possible health care worker-to-patient transmission of HIV-1 without evidence of blood exposure, we performed extensive phylogenetic analyses. The results presented here unambiguously exclude one HIV-seropositive nurse but strongly suggest another HIV-seropositive nurse as the source of transmission of HIV-1 to the patient.

Case report. At the end of May 1996, a 61-year-old HIV-negative female was admitted in a health care institution in the suburbs of Paris, France, for surgery which did not require blood transfusion. In July 1996, she suffered a severe HIV primary infection associated with hepatitis and massive weight loss. HIV-1 seroconversion occurred at that time, as evidenced by successive Western blot profiles, and was associated with high viral loads (Fig. 1). An epidemiological investigation did not identify any risk factor for HIV infection. No HIV-infected individual was found in the surgical staff, but two night shift nurses (nurse 1 and nurse 2) who had been in contact with the patient during her stay in the clinic (Fig. 1) were found to be HIV-1 seropositive. A virological investigation was requested to determine whether a nurse-to-patient contamination had occurred.

View larger version (16K): [in this window] [in a new window]   FIG. 1.   Chronology of PBMC and plasma samplings for the two nurses and the patient. Rectangles represent periods of hospitalization; vertical plain and dashed arrows indicate sampling of lymphocytes and plasma, respectively. Nurse 1 and nurse 2 had been in contact, between May 28 and June 8, 1996, each during two nights (indicated by asterisks), with the patient, who had a negative HIV-1 serology and no detectable viral RNA at the time of her hospitalization. Nurse 1 had been aware of his HIV-1 seropositivity for 5 years, and HIV-1 sequences were characterized from lymphocytes (N1) obtained on February 19, 1997, i.e., after the virological investigation was requested. Nurse 2's HIV serological status was unknown until she was diagnosed with HIV-1 and HCV infection during a hospitalization that started on June 20, 1996. The first nurse 2 HIV sequence (N2) used for molecular epidemiology analyses was from lymphocytes obtained on May 7, 1997, 1 month after introduction of antiretroviral treatment. Retrospectively, we also characterized viral sequences from cryopreserved plasma samples (JV44 and JV27) obtained during nurse 2's hospitalization. The second hospitalization (July to August 1996) of the patient was for HIV-1 primary infection. Plasma (JV32 and JV48) and lymphocyte (P) samples were obtained from the patient for viral sequence amplification. HIV-1 viral load, when available, is indicated (logarithm of the number of copies per milliliter) above corresponding sampling pointers.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Data collection. Free virus was separated from plasma samples by ultracentrifugation; RNA was extracted and reverse transcribed. DNA was extracted from peripheral blood mononuclear cells (PBMCs) obtained by density gradient centrifugation. Two reverse transcriptase (RT) gene fragments (RT1 [codons 6 to 152] and RT2 [codons 157 to 252]) were amplified from proviral DNA (PBMCs) by nested PCR (32) with external primers A35 and NE-135 as described elsewhere (31). Forward internal primers were A and B (31); reverse internal primers were 5' GGTGATCCTTTCCATCC 3' and 5' TCATTGACAGTCCAGCT 3' for RT1 and RT2, respectively. One fragment spanning the C2C4 region of the env gene encoding gp120 was amplified by nested PCR as described elsewhere (15) from both proviral DNA and viral RNA (serum). All PCR products were directly sequenced on both strands on a ABI Prism 377 automated sequencer (Applied Biosystems, Foster City, Calif.). Patient and nurse PBMC samples were collected and purified in three different hospitals (one for each individual). All plasma samples were extracted and reverse transcribed in a laboratory where HIV had never been cultured or PCR amplified. Both for the patient and nurse 2, the different cell samples (designated P and N2, respectively) were handled, DNA extracted, and PCR amplified separately. Contamination prevention procedures were strictly observed (29). Ambiguities and discrepancies in DNA-sequencing electropherograms were resolved by two independent readers. To exclude possible contamination with a known HIV strain, each sequence obtained was searched via BLAST (4) against GenBank. We also excluded the possibility of internal contamination by aligning all sequences against routine sequences obtained in the laboratory during the same period.

Sequence alignments. RT sequences were obtained by concatenation of RT1 and RT2 (618 nucleotides). Alignment was trivial (no insertion or deletion detected). Because C2C4 sequences are quite variable, we first aligned the amino acid translated sequences with Clustal W (47) using four different sets of parameters (matrix = Blossum or PAM; gap/extension penalties = 10/0.05 or 20/0.1). Positions at which the four alignments differed were excluded in the subsequent analyses (21). All alignments are available on request from M. C. Milinkovitch.

Phylogenetic analyses. (i) RT. Using different outgroup and ingroup samplings, all maximum parsimony (MP) analyses were performed with PAUP* (45) with exact branch-and-bound searches when computationally practical (i.e., in all analyses except permutation tail probability [PTP] searches and reanalyses with extended data sets). All characters were first weighted equally and treated as unordered. However, it is now well known that different types of changes can occur at different evolutionary rates, which may, in specific cases, justify differential weighting or encoding (36, 46). We therefore checked for possible saturation of nucleotide substitution types by plotting Tv versus Ti as well as the number of transitions (Ti) and transversions (Tv) versus the uncorrected pairwise distances. We also assessed the outcome of our unweighted analyses by excluding Ti or by using the Goloboff fit criterion (22) (k = 0, 2, 4, and 8). We estimated the reliability of the various inferred clades by bootstrapping (10³ to 10⁴ replicates, exact branch-and-bound searches, heuristics for extended data set), although bootstrap values (BV) may be misleading estimates of accuracy under specific conditions (36). For selected branches and using the "constraints" command in PAUP*, we computed Bremer branch supportthe number of additional character transformations necessary to collapse an internal branch (10)as an alternative to BV as an estimate of clade stability. We also performed cladistic PTP and topology-dependent PTP (T-PTP) (17, 18) analyses to statistically compare alternative phylogenetic hypotheses. Using a priori T-PTP for evaluating patient-nurse clade monophyly is valid because these tests are the results of a hypothesis of transmission formulated prior to any phylogenetic analysis.

(ii) C2C4. NJ analysis of P, N1, N2, and 29 reference sequences was performed with PHYLIP as described above. Parsimony analysis and T-PTP monophyly tests were performed with PAUP* using heuristic searches.

Nucleotide sequence accession numbers. All new sequences are deposited at EMBL under accession no. AF125604 to AF125651 for C2C4 and AF125683 to AF125806 for RT. Accession numbers of sequences used in Fig. 2 are M26727 (OYI), K03455 (HXB2), K02013 (LAI), U63632 (JRFL), M93258 (yu2), U23487 (manc), U21135 (weau), M22639 (Z2Z6), K03454 (eli), M27323 (ndk), M62320 (u455), and L39106 (ibng). Reference sequences used in Fig. 3a are the same as in Fig. 2, with the addition of AF004885, U5190, U51188, U51189, and U54771 (subtype A); AF005494 (subtype F); U88825 and U88826 (subtype G); and AF005496 (subtype H). Reference sequences used in Fig. 3b are U08794, U09127, L22939, L34667, and L22957 (subtype A); U08445, U23112, M93258, U08449, U08797, U27434, and K02013 (subtype B); U08453, U52953, and X65639 (subtype C); M22639 and K03454 (subtype D); U08810, U08456, U08458, and L03700 (subtype E); U37033, U08974, and U27413 (subtype F); U09664, U27426, and U26301 (subtype G); and U09666 and AF005496 (subtype H).

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

We compared HIV-1 strains infecting the individuals described in the case report over two regions of the viral genome: RT, and env gp120 (C2C4), representing a total of 10.7% of the whole genome. Proviral DNA sequences from lymphocytes are referred as P, N1, and N2. Phylogenies were estimated using both reference sequences and local controls (RO and JV). The stability of our results was assessed using bootstrapping, Bremer support (10), Goloboff weighting (22), Kishino-Hasegawa ML ratio tests (27), and T-PTP analyses (17, 18).

RT: patient, nurse, and reference strains. Uncorrected pairwise sequence divergence ranges from 0.16% (for comparison of HXB2 with LAI in subtype B) to 11.33% within group M and to 26.37% when group O is included. The unweighted parsimony analysis excluding group O yielded 14 MP trees (tree length = 238) whose strict consensus is shown in Fig. 2a. BV for 10⁴ branch-and-bound replicates are indicated above the branches. P is separated from N1 by three very well supported nodes (BV > 95%). All resolved nodes are stable to Goloboff weighting (22). Constraining not to keep a monophyletic [P + N2] requires a minimum of seven additional substitutions, whereas constraining a monophyletic [P + N1] requires 35 additional evolutionary events. [P + N1] was never observed in 10⁴ bootstrap replicates. A T-PTP test for [P + N2] monophyly yielded a P value of <10⁴, indicating that this clade is very unlikely the product of random data. Using the same theoretical framework, the comparison of the best trees containing a [P + N1] clade to the best trees containing a [P + N2] clade yielded very significantly better (P < 10⁴) support for the latter. The unweighted MP analysis including group O yielded trees fully compatible with the consensus shown in Fig. 2a and corresponding BV are indicated below the branches. All results under this taxon sampling were very similar to those excluding group O. 

View larger version (29K): [in this window] [in a new window]   FIG. 2.   MP and ML trees among P, N1, N2, and reference sequences (RT). (a) Strict consensus of the MP trees under no weighting of substitution types. BV under exclusion and inclusion of group O sequences are indicated above and below the branches, respectively. Including the group O sequences collapsed the branch indicated as dashed and rooted the tree (with a BV value of 53) as indicated by the arrow. N1 and P are separated by three branches (thick line) supported by BV > 95. Sequence labels correspond to reference strain names from the LANL website (see text). (b) Strict consensus of the 3 ML trees (ln L = 2062.54867). Estimation of Ti/Tv ratio, proportion of invariable sites, and gamma shape parameter for variable sites are 4.09, 0.32, and 0.64, respectively. When group O sequences are included, the branching pattern is unchanged but the tree is rooted as indicated by the arrow.

4

4

4

Envelope gp120 (C2C4): patient, nurse, and reference strains. It is recommended that strains in at least two distinct genomic regions be analyzed to determine HIV-1 subtypes (28). Analyses of C2C4 (Fig. 3b) yield results in total agreement with those described above for the RT region. NJ bootstrap, MP bootstrap, and T-PTP analyses strongly support (i) the grouping of P and N2 in a clade (BV = 99.7%, 99.7%, P_T-PTP = 5 × 10³) and (ii) the grouping of N1 with subtype G (BV = 94.1%, 71.4%, P_T-PTP = 9 × 10³).

View larger version (18K): [in this window] [in a new window]   FIG. 3.   NJ trees among P, N1, N2, and reference sequences for RT (a) and C2C4 (b). For each subtype (A to H), BV (in percent) are given for 104 NJ/104 MP replicates. Arrows indicate branches separating P from N1 with BV > 94%. Scale is in percent expected substitution per position. Note the difference of scale between RT and C2C4. Reference sequences used for RT were the same as in Fig. 2, with the addition of nine sequences from subtypes A, F, G, H (see Materials and Methods) and the exclusion of ibng (see text for details). The accession numbers of the 29 reference sequences (subtypes A to H) used for C2C4 are given in the text.

Conclusion: patient, nurse, and reference strains analyses. All RT and C2C4 analyses of patient, nurse, and reference sequences strongly support the grouping of N2 and P in a clade (i.e., a monophyletic group), to the exclusion of N1 (Fig. 2 and 3). Given that N1 is separated from P by at least three strongly supported nodes for either region, and clusters with subtype G sequences, these analyses strongly reject N1 as the source of transmission of HIV-1 to the patient. The grouping of N2 and P in a clade is consistent with the hypothesis of HIV-1 transmission from N2 to P. Alternatively, these results are also compatible with the hypothesis that P and N2 independently acquired similar geographic local variants. Hence, both for the RT and the more rapidly evolving C2C4 region, we tested the stability of the [P + N2] clade to the inclusion of local controls. Furthermore, in addition to the proviral DNA sequences obtained from N1, N2, and P lymphocyte samples used in all of the above analyses, we blindly introduced, for C2C4 analyses, viral RNA sequences obtained from the patient (JV32 and JV48) and nurse 2 (JV27 and JV44) plasma samples collected earlier (Fig. 1; Materials and Methods).

RT: patient, nurses, and 57 local controls. Fifty-five out of the 57 local control (RO) individuals were infected by HIV-1 subtype B. Sequences from the remaining two individuals were subtype A and formed, under NJ and MP analyses, a separate cluster with N1 (Fig. 4a). Local controls did not strictly cluster according to known risk factors. NJ bootstrap and MP T-PTP analyses yielded strong support for a [P + N2] clade excluding all local controls (BV = 87.1%, P_T-PTP = 6 × 10³).

View larger version (18K): [in this window] [in a new window]   FIG. 4.   Inference of phylogenetic relationships among RT sequences for P, N1, N2, and local controls. All local control sequences were obtained from plasma; P, N1, and N2 were obtained from lymphocytes from the patient, nurse 1, and nurse 2, respectively. Scale is in percent expected substitution per position. (a) NJ tree for the full set of 59 local control sequences (RO). Relevant BV (104 replicates) are indicated. Letters in bold indicate subtypes. (b) ML tree (ln L = 1913.91065) for the subset including the 15 local controls closest to [P + N2]. BV > 50% (ML, 1,000 replicates) are indicated.

Envelope gp120 (C2C4): patient, nurses, and 41 local controls. The high variability of the C2C4 region makes sequence alignment potentially problematic. Amino acid positions sensitive to alignment parameters (6 aligned positions in the C3 segment; 25 in the V4 loop) were excluded from phylogeny inference analyses (see Materials and Methods).

3

View larger version (38K): [in this window] [in a new window]   FIG. 5.   Phylogenetic relationships among C2C4 sequences from the patient, the two nurses, and multiple local controls. (a) Strict consensus among 382 MP trees (tree length = 1,191). BV obtained under unweighted MP (103 replicates) and under NJ (104 replicates) are indicated above and below the branches, respectively. na, not applicable (the group was not found in the NJ bootstrap consensus tree). (b) ML phylogram (ln L = 2662.67099) among patient, nurse 2, and a reduced set of local control sequences (i.e., redundant sequences, indicated by asterisks, were excluded from these analyses; see text for details). BV obtained under ML are indicated on the branches. Plasma viral sequences from the patient (JV32 and JV48) and nurse 2 (JV27 and JV44) were processed blind during the analyses.

View larger version (25K): [in this window] [in a new window]   FIG. 6.   ML phylogram of the [patient + nurse 2] clade. ML BV are indicated above the branches. The branching pattern of the MP bootstrap consensus tree is identical to that shown here, and corresponding MP BV are indicated below the branches. The full V4 loop sequence is indicated for each sample. Duplication of an FNSTW motif (boxed) was observed only in JV32 and JV48 (patient plasma viral sequences) as well as in the unrelated reference sequence LAI. Horizontal brackets indicate amino acid positions sensitive to alignment parameters (see Materials and Methods) when all local control C2C4 sequences are included. These positions have been excluded from phylogeny inference analyses. Note that the position indicated by an asterisk adds support to the grouping of P with JV32 and JV48.

Conclusions: local control analyses. All of our results strongly support the grouping of viral sequences from the patient and nurse 2 in a clade to the exclusion of nurse 1 and of all local controls (Fig. 4 and 5). Analyses of both RT and C2C4 corroborate the following conclusions: (i) nurse 1 is not the source of transmission of HIV-1 to the patient; (ii) it is very likely that HIV-1 was transmitted from nurse 2 to the patient or from the patient to nurse 2; (iii) the hypothesis that the transmission was from nurse 2 to the patient is strongly suggested by clinical data and consistent with the branching pattern within the [patient + nurse 2] clade.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

A virological investigation was requested by the Seine Saint-Denis sanitary authorities to evaluate the possibility of a nurse-to-patient transmission of HIV-1 in the absence of documented exposure to blood. We tested the nurse-to-patient transmission hypothesis using extensive phylogenetic analyses of viral sequences.

We could unambiguously exclude nurse 1 as the source of infection because his strain, identified as an African G subtype, was separated from the patient by several strongly supported phylogenetic nodes (Fig. 2 and 3). Our analyses of two distinct genomic regions from free viruses and integrated proviruses combined a large number of geographical and temporal local controls and consistently indicate that HIV-1 sequences from the patient and from nurse 2 share a common ancestor not shared by any other sequence analyzed (Fig. 2 to 6). This result, strongly supported by a wide range of phylogenetic approaches, indicates that HIV-1 was transmitted from nurse 2 to the patient or vice versa. However, given that nurse 2 had a full Western blot profile and a low CD4⁺ cell count (indicative of advanced infection) at the time of the patient's seroconversion, it is very likely that transmission occurred from nurse 2 to the patient.

Nurse 2 was coinfected with HCV genotype 3a (data not shown). However, no evidence of HCV infection was found for the patient either by RT-PCR or by third-generation serological tests performed 12 months later. Transmission of HIV-1, but not HCV, from coinfected individuals has already been described (9).

HIV-1 variants with an insertion of five amino acids in the V4 loop quickly emerged in the patient plasma, possibly indicating emergence of mutants escaping immunological neutralization. Insertions in the V4 loop were previously described for a case of mother-to-child transmission (11). Selection of variants during primary HIV-1 infection is common (34, 38), and our observation indicates that early samples, preferably from lymphocytes (which usually harbor more ancestral sequences [20]), should be studied when addressing transmission hypotheses.

In the near Paris Seine Saint-Denis district (1.4 million inhabitants), 9,856 AIDS cases had been reported at the end of June 1996. Among those, only 323 (3%) had unrecognized risk factors (30). In two other studies (35, 53), risk factor was not identified in 11.9 to 14.8% of patients over 50 year old. However, Stevenson et al. (43) demonstrated elsewhere that careful reappraisal of clinical data may resolve more than half of these cases. Some of the remaining cases might be associated to yet unidentified transmission routes. Although transmission of blood-borne pathogens is a major concern during health care practice (50, 51), it was suggested (26) that HIV-infected nurses would not be associated with transmission of HIV to a patient because they usually do not perform particularly invasive medical procedures. The case discussed here, indicative of atypical nurse-to-patient HIV-1 transmission, raises the issue of the mandatory screening of nurses. Because such screening would be overwhelmingly cost-ineffective to reduce these rare transmissions (12, 39, 40), suggestions that are acceptable, from both an ethical and a financial point of view, should be proposed. For example, HIV-infected health care workers, when aware of their seropositivity, could place themselves under the guidance of an expert panel which would determine, on a case-by-case basis, whether practice restrictions are appropriate (3).

Finally, utmost care is obviously advised in the divulgence of atypical HIV transmissions in a clinical context because it could lead to unnecessary alarm regarding infected medical staff (42). The case described here may be exceptional, but it indicates nonetheless that safety guidelines established for reducing risks of blood-borne transmission of pathogens between health care workers and patients should be carefully followed.