Responses to Quadrivalent Influenza Vaccine Reveal Distinct Circulating CD4+CXCR5+ T Cell Subsets in Men Living with HIV

T cell help for B cells may be perturbed in people living with HIV (PLWH), even when HIV is suppressed, as evidenced by reports of suboptimal responses to influenza vaccination. We investigated cTFH responses to the 2017–18 inactivated quadrivalent influenza vaccine (QIV) in men living with antiretroviral therapy (ART)-suppressed HIV infection who were treated in the early or chronic phase of infection, and control subjects. Here we show that seroprotective antibody responses in serum and oral fluid correlated with cTFH activation and were equivalent in all three groups, irrespective of when ART was started. These responses were attenuated in those reporting immunisation with influenza vaccine in the preceding three years, independent of HIV infection. Measurement of influenza-specific IgG in oral fluid was closely correlated with haemagglutination inhibition titre. T-SNE and two-dimensional analysis revealed a subset of CD4+CXCR3+CXCR5+ cTFH activated at one week after vaccination. This was distinguishable from cTFH not activated by vaccination, and a rare, effector memory CD4+CXCR5hiCD32hi T cell subset. The data support the use of QIV for immunisation of PLWH, reveal distinct circulating CD4+CXCR5+ T cell subsets and demonstrate oral fluid sampling for influenza-specific IgG is an alternative to phlebotomy.


Results
Study participants. Participants Supplementary Fig. S1). The influenza-A and B strains in the 2017-18 QIV had been included in at least one of the seasonal influenza vaccines from the previous three years except A/Michigan/45/2015(H1N1) pdm09 ( Supplementary Fig. S1). Individuals who had received the seasonal influenza vaccine in the past three years would therefore not have been exposed to the A/H1N1 vaccine antigen as part of a vaccination strategy.
Baseline antibody titres at Day 0, prior to vaccination, and humoral responses at Day 7 and Day 28 in sera were measured using the haemagglutination inhibition assay (HAI) 34 . More than half the cohort had a baseline HAI titre meeting the criteria for seroprotection (SP) for each antigen; 20/30 (66.7%) against A/H1N1, 21/30 (70.0%) against A/H3N2, 17/30 (56.7%) against B/Brisbane and 19/30 (63.3%) against B/Phuket (Fig. 1a). Following vaccination, the majority (86.2%) were seroprotected against all strains; only 2/14 (14.3%) HC and 2/8 (25.0%) E-HIV did not achieve seroprotection against A/H1N1, and only 1/8 (12.5%) E-HIV did not achieve seroprotection against B/Brisbane. Individuals were classified as Responders (R) to each antigen if they met criteria for seroconversion to the vaccine i.e. a ≥4-fold increase in HAI titre from baseline (Fig. 1a). There was no difference in those with E-HIV, C-HIV and HC, in the HAI titre at Day 0, 7 or 28, or the Day 28 fold change in HAI titre for any strain (Fig. 1b,c and Supplementary Fig. S2). Baseline HAI titre was higher in (Non-responders) (NR) versus R and fold change in HAI titre was higher in those not seroprotected at baseline versus those seroprotected at baseline for all strains except A/H1N1 ( Supplementary Fig. S3).
Influenza-A-specific IgG detectable in gingival crevicular fluid correlates with HAI measurements in blood and is unaffected by treated suppressed HIV infection. The human gingival sulcus is 1-3 mm in depth and lies between the teeth and the vascular periodontal tissue. It is a conduit for gingival crevicular fluid (GCF), a serum exudate which contains molecules, including antibodies, and cells from the blood 35 . We hypothesised that the humoral response to QIV would be measurable in this fluid and would correlate with the serum HAI titre. Measurements of GCF IgG against influenza-A antigens were made using an ELISA assay performed on oral GCF samples taken on Day 0 and Day 28. GCF IgG against A/H1N1 correlated with the serum HAI titre from samples taken on Day 0 (p < 0.0001) and Day 28 (p < 0.0001) (Fig. 2a). GCF IgG against A/ H3N2 correlated with the HAI titre from samples taken on Day 0 (p = 0.0152) and Day 28 (p = 0.0005) (Fig. 2b). Similar to our observations for serum HAI titre, neither the total GCF IgG nor its fold change differed between E-HIV, C-HIV and HC ( Fig. 2c and Supplementary Fig. S4). The fold changes in GCF IgG against A/H1N1 and A/ H3N2 were higher in serum responders than serum non-responders to QIV (p = 0.0004) and (p = 0.06), respectively (Fig. 2d). There was no difference in the fold change in GCF IgG for either influenza-A strain in those with or without serum seroprotection at baseline, (p = 0.94 and p = 0.59) (Fig. 2e).
Unsupervised analysis of circulating CD4 + T cells reveals different responses of three phenotypically distinct CD4 + CXCR5 + subsets. To investigate the CD4 + T cell subsets responding to QIV, we used an unsupervised computer algorithm, t-distributed stochastic neighbour embedding (t-SNE) (Fig. 3a). This resolves the high-dimensional data arising from multi-parameter FACS analysis into two dimensions, whilst preserving data structure and revealing clusters within the dataset 36 . This allowed observation of relationships in the phenotypic expression pattern of CD4 + T cells that might not be revealed through the use of hypothesis-driven 2-dimensional gating strategies. Heatmaps of the output were used to compare the relative expression of CD4 + T cell markers measured in analysis that combined all available data at each time point.
Three distinct CD4 + T cell populations differentiated by expression of CXCR5, were identified from all individuals both with and without HIV infection; P1, P2 and P3 (Fig. 3b). P1 was characterised by high expression of CD32 and CXCR5 and bore hallmarks for antigen experienced T EM , as defined by CCR7, CD45RA, CD27 and CD28 expression. Expression of CXCR3 was low in this population. Expression levels for markers of cT FH activation, including inducible T cell costimulator (ICOS) and programmed death-1 (PD-1) were medium or low 13 . P2 and P3 were more similar to one another and were dissimilar from P1, with hallmarks for T CM , as defined by CCR7, CD45RA, CD127, CD27 and CD28 expression. CXCR3 expression was high in both P2 and P3. There were dissimilarities in expression of the activation markers PD-1, ICOS and CD38 in P2 (higher expression) and P3 (lower expression) ( Fig. 3c and Table 2).
The frequency of P2 was altered following vaccination with QIV, (p = 0.0017), with an increase at Day 7 and return to baseline at Day 28 in the majority of individuals (p = 0.081 and p = 0.0012 respectively). This confirmed changes in heatmap expression density for activation markers at Day 7, including ICOS, PD-1, CD38 and CD32, visualized in P2, but not P1 and P3 (Fig. 3d-f). These CD4 + T cell populations were observed in individuals with and without HIV infection, and differences in their frequencies between E-HIV, C-HIV and HC did not reach statistical significance at each time point (Supplementary Fig. S5). Using data from the same participants, t-SNE analyses were constructed for CD19 + B cells to compare relative expression of the B cell markers CD38, CXCR5, CD27, CD32 and CXCR3 (Fig. 3g). Induction of a CD38 hi CD27 hi CXCR3 hi CXCR5 mid/lo CX32 mid/lo population of CD19 + cells was observed at Day 7 post QIV, corresponding to the population of antibody secreting cells (ASC) identified through 2-D FACS gating analysis. ASCs were distinct from the main body of live CD19 + B cells in the circulation and were clearly visible in t-SNE, appearing at Day 7. These cells were more frequent at Day 7 for some but not all individuals (p = 0.17) and returned to baseline at Day 28 post QIV (p = 0.0096) (Fig. 3h).

Hierarchical relationship of T cell populations characterised by CXCR5 expression is revealed through unsupervised analysis of the circulating T cell response to QIV. Spanning-tree Progression
Analysis of Density-normalized Events (SPADE) was used to investigate the hierarchical relationship between the CD4 + T cell populations of interest identified using t-SNE. This algorithm creates 2-dimensional trees from multi-dimensional data where the cluster size is indicative of event frequency, and colour of the median expression of each selected marker. Branching between related cell clusters allows inference of hierarchy and relationship 37 . A major branch of five cell clusters (black loop) contained populations expressing CXCR5 (Fig. 4a,b). Clusters in the major branch were most closely allied to the T CM or T EM phenotype as evidenced by expression levels of CD45RA, CCR7, CD28 and CD27 (black loop). Amongst these were two clusters, 17 and 18, most  Table S1).
Activation of circulating T-follicular helper cells correlates with fold change in HAI antibody titre following vaccination with QIV. Sequential two-dimensional gating was used to compare T cell www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ activation indicated by unsupervised analysis with humoral responses. Live CD3 + CD4 + CXCR5 + CXCR3 + cT FH were analysed ( Fig. 6a and Supplementary Fig. S7). The frequency of PD-1 + ICOS + cT FH increased significantly at Day 7 (p < 0.0001), and then returned to baseline levels at Day 28 (p < 0.0001) (Fig. 6b). There was a significant increase in the frequency of CD38 + PD-1 + ICOS + cT FH at Day 7 (p = 0.0015), that returned to baseline levels at Day 28, (p < 0.0001) (Fig. 6c). The frequency of cT FH and activated cT FH was not different between those with and without HIV infection, except for a slightly increased frequency of cT FH at Day 7 in three individuals with C-HIV ( Supplementary Fig. S8). The frequency of bulk CD4 + cells expressing CD32 was unchanged throughout the study (Fig. 6d,e). CD4 + CD32 + T cells bearing the canonical markers of cT FH (CXCR5 + ICOS + ) increased in frequency at Day 7 (p = 0.0009), and returned to baseline at Day 28 (p < 0.0001), in a response similar to that observed for other T cell activation markers (Fig. 6f) The phenotype and response of CD32 + T cells in P1-P3 was verified in 2-D gating analyses ( Supplementary Fig. S9). No consistent difference in the concentration of the soluble ligand for CXCR5, CXCL13, was observed in the serum pre and post vaccination or between participant groups ( Supplementary Fig. S10).
The CD19 + antibody secreting cell (ASC) response following vaccination with QIV was analysed in the same samples at the same time-points as the T cell response. ASCs were CD19 + CD38 hi CD27 hi (Fig. 6i and Supplementary Fig. S11). There was no difference in the frequency of ASCs between HC, E-HIV and C-HIV at any time point (Supplementary Fig. S12). The majority of CD19 + cells, median (IQR) 92.22% (88.21-94.08) expressed CD32 and CXCR5 and this did not differ between individuals with and without HIV infection ( Supplementary  Fig. S12). ASCs expressed CXCR3 consistent with a previous report 38 . Induction of CD38 hi CD27 hi ASCs was observed following vaccination with QIV at Day 7 (p = 0.0226) with resolution to baseline at Day 28 (p = 0.0005) (Fig. 6j). The frequency of ASCs expressing CXCR3 increased significantly at Day 7 (p = 0.0063) and resolved to baseline at Day 28 (p = 0.0025) (Fig. 6k). A correlation between the frequency of activated ASCs at Day 7 and the fold change in HAI titre at Day 28 was not observed for any strain (Supplementary Fig. S13).
Circulating CD8 + CXCR5 + T-follicular cytotoxic cells are not activated by vaccination with QIV at Day 7. T-SNE analysis of live CD8 + T cells, revealed a rare population expressing CXCR5 (Fig. 7a-c).
Attenuation of cellular activation and lower antibody titres post QIV in those previously vaccinated. Given the high level of pre-existing seroprotection against influenza in the cohort, the relationship between baseline HAI titre, previous vaccine exposure and cellular and humoral response following QIV was investigated. Unlike activation of cT FH at Day 7, where there was a positive relationship, baseline HAI titre correlated inversely with fold change in HAI titre for all four influenza strains ( Fig. 8a and Supplementary Table S2). Participants were categorized by their self-reported vaccination history in the preceding three years as previously vaccinated (PV), n = 17 (at least one influenza vaccination) or not previously vaccinated (NPV), n = 13 (no influenza vaccination). Thirteen of the seventeen PV (76.5%) had received it in two, or all three, of the previous three years. There was a trend for baseline HAI titre to be lower in those NPV compared with those PV, which  www.nature.com/scientificreports www.nature.com/scientificreports/ was statistically significant for influenza-B strains; B/Brisbane (p = 0.003) and B/Phuket (p = 0.0032), with no differences at Day 7 ( Fig. 8b and Supplementary Fig. S17). At Day 28 post QIV, there was a trend for HAI titre to be higher in those NPV compared with those PV, which was significant for influenza-A strains, A/H1N1 (p = 0.0126) and A/H3N2 (p = 0.0467) (Fig. 8c). The Day 28-fold change in HAI titre was significantly higher for those NPV compared with those PV for all four influenza-A and influenza-B strains (p < 0.0001) (Fig. 8d). www.nature.com/scientificreports www.nature.com/scientificreports/ Cellular activation post QIV was compared in those PV or NPV. The frequency of PD-1 + ICOS + cT FH was significantly higher in those NPV compared with those PV at Day 7, but not at Day 0 or 28 (p < 0.0001) (Fig. 8e). PD-1 + ICOS + cT FH more frequently expressed CD38 in those NPV, compared with those PV at Day 7, (p < 0.0001) (Fig. 8f). There was no difference in the frequency of cT FH at any time point between those NPV and those PV ( Supplementary Fig. S18). The fold change in the frequency of CXCR3 + ASC at Day 7 was higher in NPV compared with those PV (p = 0.0036) (Fig. 8g and Supplementary Fig. S19). Differences in the frequencies of activated www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ cT FH and activated ASCs at Day 7 post QIV were not significant, for the most part, in those with baseline SP compared with those NSP (Supplementary Figs S20 and S21). Reflecting our findings in the serum, the fold change in GCF antibody T/N ratio was higher in those NPV compared with PV for both A/H1N1 and A/H3N2 (Fig. 8h).

Discussion
In this cohort of PLWH with suppressed viraemia, we demonstrate functional immune recovery equivalent to control subjects, reflected in humoral and cellular responses to QIV, irrespective of when ART was started. Our findings support the use of QIV in immunization programmes for PLWH and the use of ART to promote recovery of vaccine-induced immunity. Humoral responses in serum and GCF were closely correlated, and were unaffected by suppressed HIV infection. This indicates measurement of influenza-A-specific IgG in GCF is a potential alternative to serum sampling that should prove robust across different patient populations.
Multi-dimensional and two-dimensional analysis of the circulating T cell response revealed distinct CD4 + CXCR5 + T cell subsets that responded differently to immunization but were unaffected by treated and suppressed HIV infection. The frequency of activated CD4 + CXCR5 + CXCR3 + ICOS + PD-1 + CD38 + cT FH (P2) increased at Day 7 following QIV in the majority of individuals, and returned to baseline frequency by Day 28 similar to previous reports 38 . A rare subset of circulating CD4 + CXCR5 + T cells that expressed high levels of CD32 (P1), was phenotypically different from cT FH , and was not increased in frequency by vaccination with QIV. Expression levels of differentiation and survival makers suggested they were highly differentiated memory cells. A third subset of CD4 + CXCR5 + T cells (P3), whilst sharing phenotypic similarities with cT FH , was hierarchically dominant but expressed persistently low levels of cT FH activation markers.
Our data support findings from several studies linking influenza vaccine-induced activation of cT FH with both the ASC response and subsequent measurements of circulating antibody 14,16,38 . The subset that responded to QIV was consistent with descriptions of cT FH; in that they were CD27 mid/hi CD127 hi CD28 hi CD45RA neg CCR7 mid reflecting a central memory phenotype, bore proteins for cellular survival signals including IL-7R and up-regulated activation markers PD-1, ICOS and CD38 at one week post vaccination. Of significance for vaccine immunogenicity, the frequency of activated cells was associated with the fold change in antibody titre and therefore with subsequent seroprotection. cT FH boost memory to influenza-specific antigen as they induce memory B cells to differentiate into ASCs 14 . CD8 + CXCR5 + cytotoxic T-Follicular cells circulated at a frequency of 1% of CD8 + T cells. They had a similar phenotype to cT FH but were not activated by QIV. Previous work indicates these cells to have potent T-cytotoxic anti-viral properties, and although not activated by QIV, could be a target for future influenza vaccines designed to induce a T cell response 18 .
The cellular response and high seroprotection rates, equivalent to healthcare control subjects, that we observed, was in individuals with fully suppressed HIV infection and good immune recovery (median CD4 count >600 cells/μl). This was regardless of interval from primary infection to treatment initiation and differs from studies reporting inferior responses to seasonal influenza vaccine in PLWH. We recruited a male cohort of PLWH, in their fifth to sixth decade of life and focused on the peak (Day 7) cellular response. Reports of a suboptimal response to trivalent vaccine in PLWH have been in cohorts where generalization to our study may be difficult, for example post-menopausal women, older people or adolescents living with HIV infection [10][11][12]21,39,40 . One study has indicated that the follicular reaction to influenza vaccine is different in PLWH, however the extent of pre-existing compromise affecting secondary lymphoid tissue from unchecked viraemia prior to starting ART is likely to vary greatly in PLWH 41 . Despite the HC being slightly younger than PLWH, responses in the three groups were equivalent. Although we did not find evidence of a weakened response to QIV in these relatively young men living with HIV infection, we cannot exclude this in older PLWH.
Unlike our findings in treated, suppressed HIV infection, recent previous exposure to seasonal influenza vaccine was associated with significantly attenuated cT FH activation after vaccination with QIV. This was not directly linked to serum HAI titres prior to vaccination, but was associated with lower titres post vaccination. Our data are in accordance with a report investigating HAI titres in healthcare personnel who undergo repeated vaccination 42 . The relationship between pre-existing antibody titre and vaccination response is complex. Some data indicate pre-existing antibody titres can negatively impact the humoral response, although this effect may be reduced in PLWH, paradoxically promoting vaccine immunogenicity 43 . In contrast, data from older people, finding pre-existing influenza-specific immunity is predictive of post vaccination antibody titres, may reflect the spectrum of immune competence in the elderly 44,45 . Our data indicate a role for cT FH in regulating the change in antibody titre with vaccination.
We observed induction of CD32 + cT FH at Day 7 in t-SNE and 2-D analyses which indicates potential functional significance for this FcγR on CD4 + T cells responding to immunisation. FUN-II is the CD32 antibody clone used in this study. It cannot distinguish between the CD32 isotypes CD32a (activating) and CD32b (inhibitory), a limitation of the recent studies using this clone to investigate HIV-DNA in CD4 + T cells 26,46 . Determination of which isotype is being expressed on the CD4 + T cells we describe should be undertaken to indicate its mechanistic significance.
We have used different analysis modalities to explore this complex dataset; each has limitations. These include inability to identify rare populations using two-dimensional analysis, inability to infer hierarchy using t-SNE and potential for natural biological variability in rare cell populations to affect clustering analysis in SPADE over time. Through a combination of these approaches we were able to tease out the specific T cell subset responding to QIV, adding to the growing body of work indicating cT FH are operational in vaccine-induced immunity. Our data identify a specific subset of cT FH that have an important function for both inducing seroprotection and controlling the antibody response and indicate that this can function optimally where HIV infection is treated and suppressed. (2019) 9:15650 | https://doi.org/10.1038/s41598-019-51961-9 www.nature.com/scientificreports www.nature.com/scientificreports/

Materials and Methods
Study enrollment. Participants were enrolled into the study during the 2017-18 Northern Hemisphere influenza season, from September 2017 until February 2018. All participants gave written, informed consent prior to enrolment. Healthcare control subjects (HC) (n = 14) receiving influenza vaccine as part of the yearly occupational health initiative, were enrolled from the staff of Imperial College London and Imperial College Healthcare NHS Trust, London UK. Male people living with HIV infection (PLWH) (n = 16), receiving influenza vaccine as part of routine care and treated for early HIV infection (E-HIV) or chronic HIV infection (C-HIV) were enrolled from the Jefferiss Wing Clinic, Imperial College Healthcare Trust, London UK. Individuals diagnosed with primary HIV infection who started ART within three months of diagnosis (E-HIV), (n = 8) were identified from a database of individuals attending a study of ART treatment of primary HIV infection (HEATHER study). Individuals attending the routine HIV outpatients clinic were classified as, HIV infection treated during the chronic phase (C-HIV), (n = 8). Participants were asked to provide details concerning their general health and lifestyle and about their influenza vaccination history over the previous three influenza seasons prior to enrolment in the study. Eleven HC (78.6%), reported a negative HIV test result within the previous 3.5 years. In these participants, using our standard flow cytometry method, the CD4 percentage of total lymphocytes was median (IQR) 30.27 (24.25-35.94). All those with HIV infection were taking ART and had an undetectable viral load (<50 RNA copies/ml) at the time of sampling, and for a minimum of six months prior to inclusion in the study. Individuals were classified as having evidence of seroprotection (SP) against each antigen if the HAI titre was ≥40. Unsupervised computer algorithms. FACS data suitable for machine learning analysis was available for 23 participants. t-SNE was performed using a FlowJo v10.4.2 plug-in. Data from all time points were gated for the sub-population desired (CD3 + CD4 + or CD3 + CD8 + or CD19 + ) by running three individual analyses. The t-SNEs were run with 1000 iterations, a perplexity of 20, Eta (learning rate) of 200 and a Theta of 0.5.
Spanning-tree Progression Analysis of Density-normalized Events (SPADE) analysis was performed using FCS express v6+ Research Edition. Data from all time points was gated for the sub-population desired CD3 + CD4 + . SPADE was run using the transformation SPADE function in FCS express with parameters of 30 clusters, maximum iterations 100, and minimum percent cells per cluster of 0.1. The parameters used in the transformation were CD8, CD4, PD-1, ICOS, CXCR3, CD45RA, CD32, CXCR5, CCR7 and CD28. The resulting output for both t-SNE and SPADE were plotted on a heatmaps and analysed for relative expression of the parameters included.
Statistical analysis. All statistical analyses were performed using GraphPad Prism software, v7.04.
Spearman's rank correlation coefficient (r s ) was calculated to compare the relationship between two independent continuous variables. Comparisons between two categorical parameters were performed using a two tailed Mann-Whitney U test. Comparisons between multiple groups were performed using a Kruskal-Wallis test and Dunn's multiple comparison test. Multiple intra-individual comparisons between time-points were performed using a Friedman test and Dunn's multiple comparisons test. For all statistical analysis, p values of less than 0.05 were considered significant and non-significant values were rounded to two significant figures. Demographics were analysed using SPSS where the percentiles were calculated using weighted average where median was defined as 50 th percentile, upper IQ as 75th and lower as 25th. Missing data was at random and was dealt with by listwise deletion for each timepoint. Two participants had missing or incomplete time-courses to perform comparative cellular experiments and were therefore excluded from analysis of cellular events. One participant had no serum available for analysis at Day 28 but data were available from the first two time points.

Data availability
All data referred to are available in the manuscript and its supplementary documents.