Temporal Trends in Prognostic Markers of HIV-1 Virulence and Transmissibility

Temporal trends in prognostic markers of HIV-1virulence and transmissibility. An observational cohort study

Nikos Pantazis PhD1, Prof Kholoud Porter PhD2, Prof Dominique Costagliola PhD3,4,5, Andrea De Luca PhD6,7, Jade Ghosn PhD8,9, Marguerite Guiguet PhD3,4, Prof Anne M Johnson PhD10, Prof Anthony D Kelleher PhD11, Charles Morrison PhD12, Prof Rodolphe Thiebaut PhD13, Linda Wittkop PhD13 and Giota Touloumi PhD1 on behalf of CASCADE Collaboration in EuroCoord

1: Athens University Medical School, Athens, Greece; 2: MRC Clinical Trials Unit at UCL, London, UK; 3: INSERM, U943, Paris, France; 4: UPMC Université Paris 06, UMR S943, Paris, France;5: AP-HP, Groupe hospitalier Pitié-Salpétrière, Service des maladies infectieuses et tropicales, Paris,France; 6: Clinic of Infectious Diseases, Catholic University of Sacred Heart, Rome, Italy; 7:Unit of Infectious Diseases, Siena University Hospital, Siena, Italy; 8: AP-HP, Hôpital Bicêtre, Service de médecine interne, Le Kremlin-Bicêtre, Paris, France;9: Faculté de Médecine site Necker, Université Paris Descartes, EA 3620, Paris, France;10: Research Department of Infection and Population Health, University College London, London,UK; 11: Kirby Institute, UNSW, Sydney, Australia; 12: FHI 360, Durham NC, USA; 13: INSERM U897 Centre of Epidemiology and Biostatistics, ISPED Bordeaux School of Public Health, University Bordeaux Segalen, Bordeaux, France

Correspondence to:

Dr Nikos Pantazis,

Department of Hygiene, Epidemiology & Medical Statistics

AthensUniversityMedicalSchool

75 M. Asias str, 115 27 Athens, Greece

Tel: 0030 210 746 2088

Fax: 0030 210 746 2205

e-mail:

Number of words: abstract (226),manuscript (3,415)

Tables (2), Figures (3), References (33)

Keywords: HIV-1, virulence, transmissibility, temporal trends, set point viral load, CD4 cell count

Abstract

Background: Measures ofpre-ART CD4 T-cell count (CD4)and HIV-1 plasma viral load (pVL) are proxies for HIV-1 virulence.Whether these proxies are changing over time has serious implications for prevention and treatment.The aim of this study is to investigate these trends.

Methods: Data were derived from the CASCADE collaboration of mainly European seroconverter cohorts. Longitudinal CD4and pVL measurements pre ART initiation or AIDS onset were analysed using linear or fractional polynomials mixed modelsadjusting for all available potential confounders.Calendar time effects were modelled through natural cubic splines.

Findings: 15,875 individuals [20·3% female; median (IQR) follow-up: 31 (14, 62) months; 8·1% drop-out rate pre-ART/AIDS)] seroconverting 1979-2008 fulfilled the inclusion criteria. Estimated (95%CI) CD4 at seroconversionfor a typical individual declined from ~770 (750-800) in the early `80s to a plateau of ~570 (555-585) cells/μL after 2002. CD4 rate of loss became faster up to 2002. Estimated (95%CI) set-point pVLincreased from 4·05 (3·98-4·12) in 1980 to 4·50 (4·45-4·54) log10 copies/mL in 2002 with a tendency of returning to lower levels thereafter. Results were similar whenrestricting analyses to various subsets,including adjusting for pVL assay, censoring follow-up at three years or using variations of the main statistical approach.

Interpretation:Our results provide strong indications of increased HIV-1 virulence and transmissibility during the course of the epidemic and a potential plateau effect after ~2002.

Funding: European Union Seventh Framework Programme.

Introduction

HIV-1 is characterized by greatgenetic diversity(1). Its continuous evolution, since the beginning of the pandemic, has given rise to many subtypes and circulating recombinant forms (CRFs)(2). The widespread use of antiretroviral treatment (ART) has also contributed to this diversity with drug resistance mutations occurring due to selective pressures of ART(3).

There is some evidence that different HIV-1 strains may differ in virulence(4).However, despite the implications for treatment and prevention, it is not yet clear whether virulence, defined as the capacity of the virus to cause disease, has changed over time. Several studies have tried to address this, but results are conflicting with some suggesting that virulence is decreasing (5-7), stable(8-10) or increasing (11-13)over time.

Older studies have used direct measures of virulence such as time between infection and AIDS/death(5-8). After the introduction of combination ART (cART) in 1995, mortality and AIDS rates have substantially decreased,rendering this approach obsolete. Thus most studies have assessed virulence through proxies such as set-pointplasma viral load (pVL), CD4 T-cell count (CD4) at seroconversion and, more rarely, CD4 declinerate (9-13). As these markers are significantly correlated with the rate of disease progressionin the absence of ART (14), higher set-point pVL, lower initial CD4 and faster CD4 loss are usually interpreted as indicative of increased HIV-1 virulence. Moreover, as the levels of pVLare associated with the risk of viral transmission(15),the study of temporal trends in set-pointpVLis of additional interest given its potential public health implications.

Using CASCADEdata (16), we attempted to estimate whether these measures have changed over the last 30 years.

Methods

Study population

CASCADE (Concerted Action of Seroconversion to AIDS and Death in Europe - is a collaboration of 28 cohorts of individuals with well-estimated dates of HIV seroconversion (seroconverters)(16). We used datapooled in September 2011 within EuroCoord ( All collaborating cohorts received approval from their regulatory or national ethics review boards (see Supplementary Appendix).

Seroconversion dates were estimated as the midpoint between the last documented negative and first positive HIV antibody test dates for the majority of participants (84·6%) withthe interval between tests being ≤3 years. For the remaining individuals, seroconversion date was estimated through laboratory methods (PCR positivity in the absence of HIV antibodies or antigen positivity with <4 bands on Western blot), or as the date of seroconversion illness with both an earlier negative and a later positive HIV test performed within a time interval of ≤3 years.All pVL and CD4 testing was performed within laboratories in industrialised countries.

Eligible individuals were those who seroconverted by31 December 2008 while≥15 years old and with≥1 CD4 and ≥1 pVL measurement available before ART initiation or clinical AIDS onset. We excluded more recent seroconverters because their follow-up would not be long enough to accurately estimate their set-point pVLor their rate of CD4 loss.Children were excluded because they are known to have different CD4 levels compared to older HIV+ individuals. SeroconvertersfromtwoAfrican cohorts werealso excluded because treatment guidelines and ART availability differ substantially in Africa (17), which results in different patterns of follow-up, censoring due to ART initiation,and AIDS onset,compared to those in resource-rich countries. Moreover, there are strong indications that CD4 levels during natural history are markedly different in HIV-1 infected individuals from Sub-Saharan Africa compared to those from Europe(18).Our analyses are thus based on data from cohorts in Western European countries, Canada and Australia.CD4 and pVL measurements taken after ART initiation or clinical AIDS were excluded from all analyses.

Statistical Methods

Exploratory data analyses regarding CD4 were based on a) available measurements within 3-6 months after the estimated seroconversion date, or b) the estimated (through subject-specific regressions) initial values and slopes. pVL exploratory analyses were based on subject-specific estimates of the set-point levels derivedthrough the available measurements taken 6-18 months after seroconversion. Longitudinal CD4 and pVL measurements were formally analyzed using mixed models withsquare-root and log10 transformations to normalize distributions, respectively. Average CD4 change was assumed linear (on the square-root scale), as it has been shown in most natural history studies,whereas for pVLa fractional polynomialwas used to capture nonlinearity as pVL levels tend to decrease more rapidly in the first months after seroconversion, stabilizing or slowly increasing thereafter (19).

Calendar time of seroconversion was allowed to affect initial marker levels and slope(s) through a 4-knot natural cubic spline. This choice was basedon visual and formal assessment of results from the exploratory analysis and the fit of similar models with varying number (2 to 8) of knots.

The list of potential confounders included sex, age at seroconversion, risk group, ethnicity, time enrolled into the constituent CASCADE cohort, and method of seroconversion determination combined with the width of the HIV test interval (> vs. ≤12 weeks). CD4 and pVL set-point (mean levels one year after seroconversion) related predictions were derived from the fitted models. Details of the statistical methods are given in the Supplementary Appendix.

Similar models were fitted to a) the subgroup of white men, infected through sex between men (MSM), b)the subgroup of individuals with a ‘midpoint’ method of seroconversion determination and a test interval <180 days and c) data with artificial censoring at 3 years after seroconversion (in order to minimize the potential effects of censoring due to AIDS, death or ART initiation). These subsets will be referred to as “white MSM”, “short HIV test interval” and “censored at 3 years” respectively.

A set of additional sensitivity analyses were undertaken:a) the pVL model was refitted including adjustments for the quantification assays, and againafter transformation of measurements into a common scale using published formulae,b) models were re-fitted after the exclusion of individuals whose region of origin was recorded as a non-industrialized country,c) alternative methods for statistical analysis of lower and higher complexity were used and d) a stratified by gender version of the main analysis has been performed.

Role of the funding source

The sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

Study population characteristics

Of 25,629 individuals in CASCADE, 9,754 were excluded from all analyses (79% of themdue to lack of the required CD4 and pVL measurements).Excluded individuals were more likely to have been infected in earlier years, to be female, infected through IDU, of non-white ethnicity and slightly younger.

The demographic and clinical characteristics of the 15,875 included individuals are shown in Table 1.The overall median (IQR) duration of follow-up was 31 (14, 62) months and 1287/15875 (8·1%) of individuals dropped-outof the study before initiating ART or progressing to AIDS.The highest proportion of women was observed in the early ‘90s and declined thereafter.Median age at seroconversion was higher in more recent seroconverters. Infection through IDU was less likely in more recent years whereas for MSM an opposite trend is apparent. Trends in the distribution of ethnicity were unclearas this information was unknown for a substantial proportion. The proportion with laboratory evidence of acute seroconversion increased over time whereas the width of the seroconversion test interval was relatively stable.

CD4 at seroconversion and rate of decline

The included individuals contributed a total of 110,168 CD4 measurements (Table 2).

Exploratory data analyses showed a negative association between CD4 and calendar year of seroconversion whereas trends of CD4 slopes were unclear.

Model-based estimates of CD4 at seroconversion according to calendar time of seroconversion are shown in Figure 1a.A clear trend for lower initial CD4 among more recent seroconverters is evident up to ~2002, followed by relatively stable levels thereafter. Estimated (95%CI) CD4 at seroconversion steadily declined from ~770 (750-800) in the early 80’s to a plateau of ~570 (555-585) cells/μL after 2002 for a typical individual (see legend to Figure 1). Assuming a linear effect of calendar time on initial CD4 for the whole study period, the estimated (95%CI) decline per year was 0·162 (0·144, 0·179) on the square root scale corresponding to between 7·5 (6·8, 8·3) and 9·1 (8·0, 10·1) cells/μL lower baseline CD4 for each subsequent calendar year (p<0·0001).Restricting analyses to the “white MSM”, “short HIV test interval” and “censored at 3 years” subsets, yielded similar results (Figure 1b, 1c, 1d).

Temporal trends for the estimated rate of CD4 decline are shown in Figure 2a. Rates of CD4 losswere relatively stable up to 1996 becoming faster between 1996 and 2002 and returning back to slowerlevels thereafter. Assuming a linear effect of calendar time of seroconversion on the CD4 loss rate, the estimated (95%CI) change in CD4 slope (square root scale) per calendar year was -0·011 (-0·016, -0·007) (cells/μL)1/2/year. This corresponds to an addition of roughly 6 (3, 8) CD4 cells/μL to the yearly rate of CD4 loss for each subsequent decade of seroconversion (p<0·0001). Restricting the analysis to the “white MSM” and “short HIV test interval” subsets yielded similar results, but without the trend for returning to less marked decay rates after 2002for the former (Figure 2b and 2c). Results from the analysis of the “censored at 3 years” (Figure 2d) subset were similar to those of the main analysis with the exception of a downward shift of the estimated CD4 slopes indicating faster CD4 loss across the whole study period.

Similar trends of both CD4 at seroconversion and CD4 loss rate were observed when the analysis was restricted to persons originating from industrializedcountries. Variations of the main statistical methods led to similar results.The stratified by sex analysis revealed that trends of CD4 at seroconversion were comparable between men and women but changes of CD4 rate of decline over time were more pronounced among women (Supplementary Appendix– Figure S1 and S2).

HIV-RNA viral load set-point

A total of 88,205HIV-RNA pVL measurementswere analyzed (Table 2).

Initial exploratory analyses revealed a trend of increasing pVL set-points among more recent seroconverters with indications for a potential plateau effect after ~2002.

Model-based estimatesof set-point pVL levels are graphically presented in Figure 3a.Estimated (95%CI) VL set-point for a typical individual (see legend of Figure 3) increased from 4·05 (3·98-4·12) in 1980 to 4·50 (4·45-4·54) log10 copies/mL in 2002.Set-point levels seem relatively stable between 2002 and 2006 but with a tendency for returning to lower levels thereafter. Assuming a linear effect of calendar time on the set-point levels, the estimated (95%CI) increase per subsequent decade of seroconversion was 0·16 (0·13, 0·19) log10 copies/mL (p<0·0001). Restricting the analysis to the “white MSM”, “short HIV test interval”and “censored at 3 years” subsets led to similar results but with slightly less pronounced changes over time for the last two cases (Figure 3b, 3c, 3d).

Adjusting for differences in pVL assays,fitting the same model to individuals from industrializedcountriesand using variations of the main statistical methods did not alter main findings. Stratifying the analysis by sex showed that changes over time in set-point pVL were similar to those found in the main analysis in both men and women. (Supplementary Appendix– Figure S3).

Discussion

Using data of HIV-infected individuals with well-estimated dates of seroconversion, we estimated that CD4 following seroconversion decreased by almost 200 cells/μL and set-point pVL increased by approximately 0·4 log10 copies/mL over the period 1979-2008.There werealso indications for an increased rate of CD4 loss during that period. These results suggest that the time required to cross the 350 CD4 cells/μL threshold decreased by up to ~50% (i.e. from a mean of7·0 years for a person who seroconverted in 1980 to 3·4 years for someone with the same characteristics who seroconverted in 2004). Given the potentially halved time to AIDS, these findings clearly indicate a significant change in the natural history of HIV infection, which had been based on data from the pre-ART era of individuals infected in the 1980s and 1990s(20).Moreover, according to the estimates of Lingappa et al.(15), our estimated increase of 0·4 log10 copies/mL in set-point pVLcorresponds to a potential 44% increase in virus transmissibility.

Findings from previous studies have been discordant(5-13). A number of reasons could have contributed to these conflicting results: most studies estimate CD4 at seroconversion and set-point pVL by averaging marker measurements within a specific time interval. These techniques may suffer from selection bias and reduced power introduced by the irregular timing of the marker measurements and their high variability within the first year after seroconversion.Many studies lacked information about seroconversion dates which obviously precludes an accurate estimation of set-point pVL and baseline CD4. Finally, diversity regarding the time periods covered by these studies, the availability and handling of crucial confounders combined with the use of differentstatistical methods could, at least partly, explain the variability of the results.A recent meta-analysis (21)examined patterns of baseline CD4 (at seroconversion, if available, otherwise at HIV diagnosis) and set-point pVL. Even though most of the included studies were of individuals with unknown dates of HIV seroconversion, pooled estimates from this meta-analysis, extrapolated over a 30-year period, were close to our own.

The current study is, to our knowledge, the largest seroconverter study exploring this issue, has very wide calendar times of seroconversion coverage, and uses a unified approach to explore temporal trends in the proxies of interest. The models we used included adjustments for several factors but the available data do not include all cofactors that potentially modulate disease progression and thus, as in all observational studies, residual confounding cannot be ruled out.

A potential source of bias in our study is the diversity of assays and platforms used to quantify pVL and CD4 over the course of the epidemic across the different collaborating cohorts. However, the evolution of viral load assays has targeted lower thresholds of quantification while, in our study, the vast majority of measured pVLs were at much higher levels where the concordance between assays is high (22). Moreover, according to a study which compared older with modern, more sensitive assays (23) set-point pVL measurements by the former were slightly higher which, if true, would mean that our estimates are conservative. Thus, a systematic shift over time towards overestimation of pVL seems an unlikely explanation for our findings. Relevant sensitivity analyses yielded practically the same results as the main analysis. Similarly, a systematic bias towards underestimation of CD4 in more recent years does not seem likely either.

Deterioration of HIV-RNA in stored samples is another potential issue: HIV-RNA quantification has been introduced into clinical routine mainly after 1996, thus most of the pVL measurements for the initial years of the epidemic were performed retrospectively on frozen samples. However, the increasing trends in set-point pVL, observed in our study, continue after 1996 and the estimates of the deterioration rate in frozen samples are mild (24) and are not enough to fully explain the pre-1996 changes.

Another potential issue is the high percentage of participants with missing information on HIV-1 subtype (72%) or ethnicity (54%) in our data as it is known that these factors may influence baseline CD4 and pVL levels and their evolution (18, 25). The increasing proportion of non-white individuals (mainly of African ancestry), infected with non-B HIV-1 subtypes along witha likely changingsocioeconomic profile of our study populationcould have some influence on our results. However, when we refitted our models restricting to white MSM, a group most probably dominated by subtype B infection, or to individuals originating from industrialized countries, we found no major deviations from our main findings.