For the current study 101 participants were selected from a larger clinical study (n = 727) that compared the serological responses to traditional egg-based [standard-dose IIV (Fluzone^® Quadrivalent by Sanofi Pasteur or Fluarix Quadrivalent by GSK, 15 μg of HA per strain)] vs egg-free influenza vaccines [ccIIV (Flucelvax™ Quadrivalent by Seqirus, Inc., 15 μg of HA per strain) or RIV (Flublok^® Quadrivalent by Sanofi Pasteur, 45 μg of HA per strain)] during 2018–19 influenza season in the repeatedly vaccinated HCP [23]. The 101 participants provided blood sample for additional testing for peripheral blood mononuclear cells (PBMC) to assess cell-mediated immune (CMI) responses and the focus of the current paper is B cell and T cell response to vaccination. Among the 101 participants in this study, the demographics across the three vaccine arms-ccIIV(n = 37), IIV(n = 41), and RIV(n = 23), were comparable in terms of age and gender. Specifically, 86–88 % of participants were female, and 49–52 % were within the 45–64 age range (Table 1).

To quantify HA-specific B cell responses to the vaccines via flow cytometry, we employed fluorescently labeled recombinant trimeric H1 and H3 HA (HAΔSA) and HA stem probes and validated them prior to their use (please see methods section). To assess HA-specific B cell responses to the vaccines, we focused on IgG+ B cells, which represent the majority of class-switched memory B cells in circulation, play a crucial role in recall responses [44] and are essential for the protection conferred by inactivated influenza vaccines [28].

The vaccine strains included in the 2018–19 influenza vaccines were A/Michigan/45/2015 (H1N1)-like, A/Singapore/INFIMH-16–0019/2016 (H3N2)-like, B/Colorado/06/2017 (Victoria)-like, and B/Phuket/3073/13 (Yamagata)-like [45]. We used the multiplexing approach to study the responses to all components of the quadrivalent vaccines (H1N1, H3N2 and both IBV lineages), when there was limitation of the clinical material and need to interrogate the high number of cells for detection of low frequency HA-specific cells. The individual immune response to each influenza vaccine component varies markedly based on host-specific factors—particularly immunological history with influenza viruses, pre-existing antibody titers, and demographic characteristics—as well as vaccine-specific properties including antigen design. When optimal conditions converge, B lymphocyte activation triggers two critical protective mechanisms: the rapid differentiation of antibody-secreting cells (ASCs) that produce class-switched, high-affinity neutralizing antibodies providing immediate protection, and the establishment of durable memory B cell populations that enable accelerated responses to subsequent viral encounters. We studied HA-specific B cell response in the context of serological immunity specific for each vaccine component.

Pre-vaccination hemagglutination inhibition (HI) geometric mean titers (GMTs) against the H1N1 cell-grown virus were as follows: ccIIV (27.4, 95 % CI: 16.5–45.7), IIV (46.6, 95 % CI: 29.4–73.7), and RIV (64.6, 95 % CI: 37.7–110.7) (Table 2; Fig. 1A). At one-month post-vaccination (m1), RIV GMTs (147.3, 95 % CI: 101.8–213.2) were significantly higher than those of ccIIV (47.6, 95 % CI: 27.8–81.3) and IIV (61.7, 95 % CI: 38.7–98.5), with statistical significance specifically for RIV versus ccIIV. No significant differences in GMT fold rise (FR) were observed among the vaccine groups at m1 or month six (m6) (Table 2; Fig. 1B). At m1, RIV induced a higher H1 HA-specific IgG+ MBC response (FR 2.5, 95 % CI: 1.7–3.3) compared to IIV (FR 1.5, 95 % CI: 1.2–1.7), with P = 0.004 (Fig. 2A). The responses to Group 1 HA stem-specific B cells were comparable across all vaccine types (Fig. 2B).

Pre-vaccination microneutralization (MN) GMTs against the H3N2 cell-grown virus showed no significant differences among the vaccine arms: ccIIV (72.9, 95 % CI: 49.7–107.0), IIV (112.1, 95 % CI: 80.4–156.4), and RIV (90.7, 95 % CI: 55.3–148.8). At m1 and m6 post-vaccination, RIV GMTs were significantly higher than those of ccIIV or IIV (Table 2; Fig. 1A). The GMT FR from pre- to post-vaccination for H3N2 was 8.5 (95 % CI: 4.1–12.9) for RIV, while it was 3.5 (95 % CI: 1.3–5.7) for ccIIV and 1.2 (95 % CI: 1.0–1.5) for IIV (Table 2; Fig. 1B). The RIV group consistently demonstrated a greater induction of H3N2 HA-specific B cells compared to ccIIV and IIV at all time points, except for the RIV versus IIV comparison at day 7 (d7). Specifically, for RIV, the fold rise in cell percentage was 1.6 (95 % CI: 1.2–2.1) at d7, 1.7 (95 % CI: 1.3–2.2) at m1, and 1.4 (95 % CI: 1.1–1.7) at m6 (Fig. 2A). In contrast, ccIIV and IIV showed only slight increases in HA-specific B cells (1.2 with 95 % CI: 0.1–1.4 and 1.3 with 95 % CI: 1.1–1.4, respectively) at d7 post-vaccination. There were no significant differences in B cell responses to Group 2 HA stem among the vaccine groups (Fig. 2B).

Pre-vaccination hemagglutination inhibition (HI) geometric mean titers (GMTs) against the B(Victoria) cell-grown virus were similar across the vaccine arms: ccIIV (49.3, 95 % CI: 33.8–71.9), IIV (72.0, 95 % CI: 51.3–100.9), and RIV (60.1, 95 % CI: 45.3–79.6) (Table 2; Fig. 1A). At one-month post-vaccination (m1), GMTs remained comparable among the groups. However, by month six (m6), ccIIV GMTs (17.6, 95 % CI: 10.9–28.4) were significantly lower than those of IIV (50.4, 95 % CI: 35.9–70.8) and RIV (57.5, 95 % CI: 39.9–82.8). Despite these differences in absolute titers, the fold rise (FR) in GMTs from pre- to post-vaccination was comparable across all vaccine types at both m1 and m6 (Table 2; Fig. 1B).

Notably, RIV induced a higher level of B(Victoria) HA-specific B cells at m1 (FR 1.9, 95 % CI: 1.6–2.2) compared to ccIIV (FR 1.6, 95 % CI: 1.4–1.8) (Fig. 2A). In contrast, the frequency of IBV stem-specific B cells in the ccIIV group decreased at d7, m1, and m6 compared to baseline levels (Fig. 2B). Both IIV and RIV demonstrated low-level induction of IBV stem-specific B cells, with a significant difference observed between the IIV and ccIIV groups at m1.

Pre-vaccination HI GMTs against the B(Yamagata) cell-grown virus were similar across the vaccine arms: ccIIV (98.8, 95 % CI: 71.2–137.2), IIV (100.8, 95 % CI: 73.1–139.0), and RIV (81.3, 95 % CI: 59.9–110.4) (Table 2; Fig. 1A). GMTs and FRs at both m1 and m6 were comparable among the three vaccine types (Table 2; Fig. 1A & B). However, RIV induced a significantly higher level of B(Yamagata) HA-specific B cells at m1 (FR 2.3, 95 % CI: 1.8–2.9) compared to ccIIV (FR 1.4, 95 % CI: 1.1–1.7) (Fig. 2A).

Overall, RIV induced higher HA-specific B cell response and/or HI or MN antibody response in HCP compared to ccIIV and IIV. The magnitude of RIV response was dependent on the vaccine component. H3N2 response demonstrated much higher difference between RIV vs inactivated vaccines (ccIIV and IIV) in B cell and antibody responses. H1N1-specific response followed the pattern of H3N2-specific response but with less significant response of RIV. The HA-specific B cell induction was higher for RIV, however, HI response to IBV components of vaccines was comparable among the vaccine groups. Additionally, we assessed the response of circulating T follicular helper (cTfh) cells (CXCR5+ PD1+ CD4+ B cells) to vaccination, which reflects the overall immune response to all vaccine components (Supplementary Fig. 1B; Supplementary Table 1B). Consistent with the B cell responses findings, RIV elicited a higher cTfh cell response at m1 (FR 1.4, 95 % CI: 1.1–1.8) compared to IIV (FR 1.0, 95 % CI: 0.8–1.1) and ccIIV (FR 1.0, 95 % CI: 0.8–1.2), with the difference reaching statistical significance when comparing RIV to IIV (Fig. 3).

Using the HA probes we focused on memory recall B cells expressing HA-specific antibodies and examined the correlation between MBC, cTfh and serological responses. In the RIV group, d7/d0 H3 HA-specific B cell FR correlated with H3N2-specific MN GMTs at m1 and m6 (Fig. 4A–B) (p = 0.0006, r = 0.67 and p = 0.0006, r = 0.72), although lower MN titers were observed at m6. Additionally, d28/d0 H3 HA-specific B cell FR correlated with H3N2-specific MN GMTs at m1(p = 0.018, r = 0.05). However, no correlations were found between B cell FR and GMTs for other vaccine groups or components, as well as no correlation between cTfh and B cell/serological responses was detected (data not shown).

This study included a subset of participants from a larger randomized comparative immunogenicity trial evaluating the recombinant influenza vaccine (RIV), cell culture-derived inactivated influenza vaccine (ccIIV), and conventional egg-based inactivated influenza vaccine (IIV) among healthcare personnel (HCP) aged 18–64 years (NCT03722589) [23]. Participants who consented to the main trial were randomized by age stratum (18–49 years and 50–64 years) to receive a single dose of the Northern Hemisphere RIV (45 μg of HA per strain, Flublok™ Quadrivalent by Sanofi Pasteur), ccIIV (15 μg of HA per strain, Flucelvax™ Quadrivalent by Seqirus, Inc.), or IIV (15 μg of HA per strain, either Fluzone^® Quadrivalent by Sanofi Pasteur or Fluarix Quadrivalent by GSK) during the 2018–19 influenza season. Following randomization, participants were invited to enroll in this sub study, which included peripheral blood mononuclear cell (PBMC) collection, until we reached a total of 101 participants, evenly distributed across age strata and vaccine arms (Table 1).

Serum samples were collected just prior to vaccination and at approximately one, and, six months post-vaccination to assess serological immune responses. Whole blood was also collected pre-vaccination and at approximately seven days, one month, and six months post-vaccination to evaluate humoral and CMI responses via flow cytometry. PBMCs were isolated using the BD Vacutainer CPT method (BD Biosciences) and cryopreserved until analysis.

MN assay was performed against H3N2 virus only as HI assay became unreliable due to altered binding characteristics of receptor binding site to the red blood cells. The assay has been described previously [23,34]. Microneutralization assays were performed by Southern Research Institute using CDC protocols after technicians had passed CDC proficiency panels. Sera were heat inactivated at 56 °C for 30 min. Two-fold serial diluted sera were incubated with 100 tissue culture infectious dose per 50 μl of A/Singapore/INFIMH-16–0019/2016(H3N2) virus and incubated at 37 °C 5 % CO2 for 1 h. The virus-sera mixture was used to infect Madin-Darby Canine Kidney (MDCK) cells, incubated for 18–20 h at 37 °C with 5 % CO2, and then fixed using cold 80 % acetone in phosphate buffered saline. An ELISA using influenza A nucleoprotein monoclonal antibodies was used to quantify influenza viral proteins. MN titers were defined as the reciprocal of the highest dilutions of serum that resulted in 50 % virus neutralization. The virus was propagated in MDCK-SIAT1 (Millipore Sigma, USA, Cat 05071502). The limit of detection was ≥10 [55].

The assay has been previously described [23,56]. Hemagglutination inhibition assays were conducted at Battelle Eastern Science and Technology Center, Aberdeen, MD using CDC protocols after technicians had passed CDC proficiency panels for H1N1 and B viruses. Serum samples were treated with receptor destroying enzyme, tested for non-specific agglutinins and adsorbed as needed using washed 0.5 % turkey erythrocytes and cell-grown strains A/Michigan/45/2015 (H1N1), B/Colorado/06/2017 (Victoria lineage), and B/Phuket/3073/2013 (Yamagata lineage) propagated in MDCK-SIAT1 (Millipore Sigma, USA, Cat 05071502). Two-fold serial dilutions were performed starting at 1:10 and the HI titer was defined as the reciprocal of the last dilution of serum that completely inhibited hemagglutination. Serum from all time points was tested on the same plate. All samples were tested in duplicate and the final titer was the geometric mean of the duplicate titers. Samples with results <10 were assigned a titer of 5. If initial duplicate titers were not within one two-fold serial dilution of each other, all test samples from the same individual were repeated in duplicate. The limit of detection was ≥1:10 dilution of sera [57]. HI assays were performed against H1N1 and IBVs.

The rec HAs expressed as RBS-mutated hemagglutinin (ectodomain or H1 domain) or HA stem domain were codon optimized and produced in the baculovirus-insect cell expression system as trimers (with C-terminal foldon and His-tag) as previously described ([58] (Supplementary Table 2). Secreted soluble trimeric rec HA protein was recovered from the cell culture supernatant by tangential flow filtration through a 30 kDa (HA ectodomain or HA1 domains) or 10 kDa (HA stems) molecular weight cut-off membrane, metal affinity chromatography, and gel filtration chromatography.

Rec H1N1 HAs with Tyr91Phe (Y91F) substitution ([59]) and rec H3N2 HAs with the equivalent Tyr98Phe (Y98F) substitution were expressed as ectodomain to generate HAΔSA probes [42,60–62]. For the IBV HAs we performed a structural analysis with various residues selected for mutation. We expressed and purified B/Colorado/06/2017 constructs and assessed their glycan binding. Microarrays were printed and supplied under a contract with James Paulson at the Scripps Research Institute (La Jolla, CA). We prepared rec HA-antibody pre-complexes of wt and F95A IBV proteins using IBV HA1, a primary mouse anti-penta-His antibody (Qiagen), and a secondary anti-mouse-IgG-Alexa Fluor antibody (ThermoFisher Scientific) in a 4:2:1 M ratio. After 1 h of incubation on ice, we subjected these prepared mixtures to glycan array analysis as described previously ([63]). We found that the F95A substitution abrogated binding to glycan microarrays (Supplementary Fig. 2), and we also introduced it into the B/Phuket/3073/13 construct. We expressed the rec IBV HAs as HA1 trimers.

We designed A/Michigan/45/2015 (group 1) HA stem and A/Singapore/INFIMH-16–0019/2016 (group 2) HA stem based on the GEN-4 construct [64]. We labeled 500 μg of each protein following the manufacturer’s protocol. We chemically labeled the probes at primary amines with DyLight dyes 488, 550, 650 (ThermoFisher Scientific, USA, Cat 53024, 84530, 84535) or PerCP-Cy5.5 (Abcam, USA, Cat ab 102911).

We used a single IBV HA stem probe because of the similarity between the HA stems of Victoria and Yamagata lineages. We prepared an IBV HA stem probe from the B/Brisbane/60/2008 (Victoria lineage) HA-based stem construct BHA10F (Supplementary Table 2) that was based on an earlier design for an H1 mini-stem [65]. We used the native connectivity of HA for the design, and the construct BHA10F contained residues 151–361 of HA1, a four-residue GSAG linker, followed by 3021–3341 of HA1, a seven residue GSAGSAG linker followed by 3842–4632 stretch of HA2 (rcsb.org/structure/4fqm). We incorporated previously characterized [66], low-pH conformation destabilizing mutations, M4172D and L4202D, into the design. To promote the formation of a trimer, BHA10F had the globular, β-rich trimerization motif ‘foldon’ [67].

The E. coli codon-optimized BHA10F were synthesized (GenScript, USA) and cloned into the expression vector pET-28a (+) (Novagen) between NdeI and HindIII restriction sites. The protein sequence was in-frame with the N-terminal His-Tag from the vector pET-28a (+). Gen-Script USA, Inc. prepared rec IBV HA stem protein by over-expressing in E. coli cells and obtaining from the soluble fraction of the cell lysate using Ni-NTA resin. Endotoxin was measured (4.6 EU/mg (LAL Endotoxin Assay Kit, GenScript, USA). We stored the protein in a buffer (50 mM Tris-HCl, 10 % glycerol, 150 mM NaCl, pH 8.0) at —80 °C. 500 mg of protein was chemically labeled at primary amines with DyLight dyes 550 (ThermoFisher Scientific, USA, Cat 84530) following the manufacturer’s protocol.

To study the effect of RBS mutation on the antigenic characteristics of rec HAs, we immunized 6–8-week-old female BALB/c mice (Jackson Laboratories, Bar Harbor, ME) with A/California/07/09(H1N1) MIV (GSK, 10 μg/mouse) or HA wt or HA Y91F (HAΔSA) (10 μg, and 1 μg of HA per mouse), as well as PBS (negative control). We immunized animals four times every four weeks. We tested the response by HI against the A/California/07/09 virus three weeks after each vaccination. We purified splenocytes four weeks after the last vaccination by manually grinding the spleen in the complete Roswell Park Memorial Institute (RPMI) medium. We passed the cells through a cell strainer, treated them with 2 ml RBC lysis buffer for 3 min, and neutralized with up to 30 ml of total volume with RPMI supplemented with 10 % Fetal Bovine Serum (FBS). Finally, the cells were collected, resuspended in PBS to detect HA-specific B cells by flow cytometry. To study the binding of sera from MIV-vaccinated mice to Fluorescent-tagged or unaltered HAs, we immunized BALB/c mice with A/California/07/09 MIV(H1N1) MIV (GSK, 10 μg/mouse) or A/Texas/50/2012(H3N2) MIV (GSK, 10 μg/mouse), boosted four weeks later and sera were collected three weeks post-booster immunization for ELISA analysis.

The HAΔSA probes (H1 HA Y91F, H3 HA Y98F and IBV HA F95A Victoria and Yamagata) were validated by different experimental approaches using human PBMCs as well mouse model. We performed flow cytometric analysis of PBMCs stained with wt HA or mutated HA proteins. The RBS mutations effectively minimized non-specific binding of HA probes to B cells (Supplementary Fig. 3). To demonstrate that the RBS Y91F mutation did not alter antigenicity of rec HAs, BALB/c mice were immunized with wild type (wt) HA, HAΔSA or monovalent influenza vaccine (MIV) of the same viral strain (as positive control) and PBS as negative control. The serological response measured by HI assay and HA-specific B cell response were comparable between wt and mutant HAs (Supplementary Fig. 4). The results were consistent with those of the previous publications [42,62] which demonstrated that H3N2 HA Y98F and the corresponding H1N1 HA Y91F mutations did not alter the antigenicity of rec HA. The HA probes were chemically labeled with fluorescent DyLight dyes. To ensure that fluorescent dye conjugated to HA did not change antigenicity of rec HAs, we compared binding of sera from mice vaccinated with H1N1 or H3N2 to conjugated HAΔSA vs unconjugated HAΔSA by ELISA. There was no difference in the binding based on conjugation status (Supplementary Fig. 5). Hence, fluorochrome conjugation did not impact HA protein antigenicity.

For detecting HA stem-specific B cells, we utilized fluorescently labeled group 1 and group 2 HA stem probes [64]. These HA stem probes were tested by Bio-Layer Interferometry (BLI) for binding with the previously characterized stem-specific cross-binding monoclonal antibodies (mAbs): CR6261 (group 1 HA stem- specific) [68], CR8020 (group 2 HA stem - specific) [69], and CR9114 (group 1 and group 2 HA stem-specific) [70] and Fi6v3 (group 1and group 2 HA stem specific) [71] (Supplementary Fig. 6). HIV-specific VRC-01 mAb [72] was used as negative control. The HA stem probes demonstrated the expected affinity, binding only the relevant mAbs. Due to the structural similarities between the HA stems of the Victoria and Yamagata lineages a single HA stem probe was used for detection of IBV HA stem-specific B cell response (see methods section). Testing for probe specificity through dual probe staining (using two HAΔSA of the same specificity conjugated to different fluorochromes) [73] revealed significantly higher percentage of double positive HA-specific cells in IgG+ B cell subset (72.7, 95 %CI: 41.9–103.5) compared to naïve CD27-IgG- B cells (28.5, 95 % CI 5.1–51.9) (Supplementary Fig. 7), consistent with the findings of Koutsakos et al. [42].

Dual HAΔSA probe staining [73] confirmed the specificity of the corresponding HA probes used for staining the clinical samples, with 76–90 % of double-stained IgG+ B cells for H1N1, H3N2, and IBV HAs, and 69–88 % for the HA stem probes (Supplementary Fig. 8). Two antibody panels were developed for staining HA-specific IgG+ B cells: one panel included H1N1, H3N2, and both IBV HAΔSA probes (Panel 1), and the second panel included HA stem probes (group 1, group 2, and both IBV HA stems) (Panel 2) (Supplementary Fig. 1, Supplementary Table 1 A).

Cryopreserved PBMCs were thawed, counted, and stained for flow cytometric analysis. The HA-specific B cell response was assessed through co-staining with recombinant RBS-mutated hemagglutinin probes (HAΔSA) derived from the 2018–19 influenza vaccine viruses (using cell staining panel 1) and HA stem probes (using cell staining panel 2) (Supplementary Table 1 A; Supplementary Table 2; Supplementary Fig. 1 A). T cell responses were evaluated using multi-parametric flow cytometry, including markers for circulating T follicular helper cells (cTfh, CXCR5+ PD1+ CD4+) (Supplementary Table 1B; Supplementary Fig. 1B). The target cell numbers were 2–3 million for each B cell panel and 1–3 million for T cell panel.

Data acquisition was performed using either a BD LSRFortessa or a BD FACSymphony A5, with analysis conducted using BD FACS Diva Software and FlowJo v9. B and T cell responses were presented as fold rise (FR) in the percentage of cells at each time point compared to baseline levels at day 0. To ensure the quality control, we set a cut-off of fewer than 5 HA-specific cells for excluding samples from analysis, addressing potential outliers arising from factors such as the rarity of HA-specific memory B cells in baseline samples, inadequate cell numbers, and variability in HA probe specificity and conjugated fluorochrome.

We coated the Immulon 2 HB 96-well microtiter plates (Thermo Fisher Scientific, USA) with 100 μl of 5 μg/mL labeled or unlabeled A/California/07/09(H1N1) or A/Texas/50/2012(H3N2) rec HAs in phosphate-buffered saline (PBS) at 4 °C overnight. We washed the plates with wash buffer (PBS-Tween 0.05 %) and blocked them with 4 % bovine serum albumin (BSA) in wash buffer for 1 h at 37 °C. We added a 10-fold serially diluted mouse serum of the corresponding specificity to the plates. After 2 h of incubation at 37 °C, we washed plates and added anti-mouse IgG conjugated to HRP (Invitrogen, USA, Cat 62–6520) at 1:1000 in 100 μl wash buffer. After 1-h incubation at room temperature, we washed the plates. We added 50 μl TMB substrate solution (Thermo Scientific™, USA, Cat ENN301) to wells followed by 50 μl of TMB stop solution (Thermo Scientific™, Cat N600) and immediately read at 450 nm on Bio Tek Synergy 4 (Agilent).

We propagated influenza A/California/07/2007(H1N1) stock in the allantoic cavity of 10-day-old embryonated chicken eggs. We pooled allantoic fluid, clarified by centrifugation, sequenced, titered, aliquoted, and stored at –80 °C until use.

We treated the sera collected from immunized mice with receptor-destroying enzyme (RDE, Denka Seiken) at 37 °C overnight. We tested sera against influenza viruses by the standard HI assay using 1 % turkey red blood cells (RBCs). Briefly, we added 25 μl of 1× PBS to wells of a 96-well V-bottom plate (Corning). Next, we added 50 μl of RDE-treated sera to each column and then serially diluted. Then, we added 25 μl of 4 HA units of influenza virus to each well and incubated at room temperature (RT) for 30 min. Finally, we added 50 μl of 1 % standardized turkey RBCs in PBS to each well and incubated at RT for exactly 30 min. We recorded the reciprocal serial dilutions of the sera that showed complete inhibition of hemagglutination as the HI titer.

The Octet R8 system (Sartorius, NY, USA) was used for to measure binding kinetics of mABs to group 1 or group 2 HA stem rec proteins. HIigK biosensors (Sartorius, Cat 18–5122) were equilibrated with Octet Kinetic Buffer [KB] (Sartorius, Cat 18–1105) for 10 min and loaded with 10 μg/ml of HA stem protein for 200 s followed by baseline in KB for 140 s. The association with 30 μg/ml of mABs was performed for 100 s following by dissociation for 300 s. For negative control, HIV-specific VRC-01 mAb [72] was used. Data were analyzed with Octet Analysis Studio (Sartorius, NY, USA).

Statistical analyses were performed using GraphPad Prism. We utilized one-way ANOVA with Tukey’s multiple comparisons test and paired t-tests as indicated in the figure legends. Results are reported as geometric means or means with 95 % confidence intervals (CI), and correlation was assessed using Pearson correlation coefficients (r).