Cytokine levels were measured on a total of 280 subjects; 151 overlapped with our previous study in which we measured IFN-induced gene expression (Table 1) [5]. Most of the study subjects were female and of Caucasian ethnicity, with a greater proportion of males and a lower proportion of Caucasians in the ANA−HC, as compared to the other ANA+ participant subsets. Fifty-five of the ANA+ individuals lacking a SARD diagnosis that were followed longitudinally had completed 2 years of follow-up at the time of the study, 13 of which developed new SARD criteria during this period. The new clinical criteria that developed and resulting diagnoses for the progressors are outlined in Additional file 1: Suppl. Table 1. For both progressors and non-progressors, the majority of individuals were female and Caucasian. Progressors had significantly more autoantibody specificities, as measured by Bioplex, than non-progressors and were more likely to be of South Asian ancestry. The proportion of UCTD patients within the subset of ANA+ individuals lacking a SARD diagnosis that demonstrated progression (37.7%) was not significantly different from that in non-progressors (53.8%, p = 0.33), with similar clinical features being present at baseline in both groups (Additional file 1: Suppl. Table 2) and similar proportions of individuals receiving antimalarial therapy (23% in progressors and 16.7% in non-progressors, p = 0.68).

Table 1 Study participant characteristicsIFN-induced gene expression and several of the cytokine measures were elevated in patients with a SARD diagnosis

Figure 1 shows the levels for the various assays used to assess IFN or its effects in all study subjects. As previously shown by ourselves and others, peripheral blood IFN-induced gene expression, as determined by the IFN5 score, was significantly elevated in SARD patients and in ANA+ individuals lacking a SARD diagnosis, although not to the same extent as that seen in SARD. IFN-α, as measured by a high-sensitivity ELISA, was also significantly elevated in SARD patients. However, for a considerable number of study participants, the IFN-α levels were below the lower limit of detection of this ELISA. To better capture differences in this lower range, we measured serum IFN-α by Simoa. As expected, this improved the ability to detect IFN-α in more subjects, permitting the identification of elevated levels of IFN-α not only in SARD patients, but also in both subsets of ANA+ individuals lacking a SARD diagnosis, relative to ANA−HC. In contrast to these more direct measures of type I IFN, the levels of CXCL-10 did not differ from ANA−HC in any of the ANA+ groups and significantly elevated levels of Galectin-9 were only seen for SARD patients. Similar findings were also obtained for IFN-γ, which was not elevated in any of the ANA+ groups.

Fig. 1

Cytokine levels in the ANA+ participant subsets. Scatterplots showing the results for the IFN5 score and the cytokines, IFN-α measured by high-sensitivity ELISA, IFN-α measured by Simoa, CXCL-10, Galectin-9, and IFN-γ (all shown using logarithmic scales). From left to right are shown results for healthy controls (ANA−HC), asymptomatic ANA+ individuals (ANA+NS), undifferentiated connective tissue disease (UCTD) patients, and systemic autoimmune rheumatic disease (SARD) patients. Results to the right of the dashed line are those for the individual SARD conditions, SLE, SS, and SSc. Each circle represents a single subject, with the bars indicating the median for the subjects and error bars denoting the interquartile range. Significant differences between ANA−HC and the ANA+ groups are indicated with asterisks, *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, and those between SARD and the other ANA+ subsets indicated with cross hatches, using the same scale

Given the large differences in age between some of the groups, we performed a multivariable analysis incorporating age as a predictor to determine the impact it had on the results, and for cytokines that were significantly impacted by age, we did an age-adjusted analysis of the statistical differences between groups. The only cytokine measurements that varied significantly with age were IFN-α measured by ELISA (p=0.002) and Simoa (p=0.021), where an age-adjusted analysis led only to a loss of significance between ANA−HC and ANA+NS or UCTD for Simoa, but no other comparisons between groups.

We also assessed whether the duration of storage at −80° C affected the levels of cytokines detected by performing a linear regression analysis and found no association between the age of the sample and cytokine levels.

IFN-α is a good surrogate marker of the IFN signature, whereas CXCL-10 and Galectin-9 reflect both type I IFN and IFN-γ elevations

To further explore the interrelationship between the various surrogate markers of IFN-induced gene expression and the IFN5 score, as well as each other, we examined their correlation in all study subjects (Table 2). There was a moderate to strong correlation between the IFN5 score and IFN-α measured by ELISA or Simoa, which was strongest for Simoa. In contrast, there was only a weak correlation between the IFN5 score and serum levels of CXCL-10, Galectin-9, and IFN-γ.

Table 2 Correlations between cytokines for all subjects

Given the discordance between our results and those previously reported in the literature for predominantly SLE patients [15], we questioned whether the correlation between these cytokines and the IFN5 score varied based on the diagnosis. In ANA+NS, there was no correlation between either CXCL-10 or Galectin-9 and the IFN5 score, whereas there was a weak correlation for Galectin-9 with the IFN5 score in UCTD and SARD patients (ρ = 0.29 and 0.25, respectively, both p ≤ 0.05). In contrast, there was a moderate to strong correlation between IFN-α, regardless of how it was measured, and the IFN5 score for all 3 ANA+ groups (ρ = 0.52–0.71, all p < 0.0001).

Previous reports have suggested that CXCL-10 and Galectin-9 are predominantly driven by IFN-γ [20, 21]. When all subjects were included, there was a weak correlation between these two cytokines and IFN-γ, which approximated that seen for the IFN5 score and IFN-α, with no marked differences observed between ANA+ individuals with and without symptoms. These findings raise the possibility that the production of these cytokines in ANA+ individuals is not solely induced by IFN-γ. Nevertheless, the levels of CXCL-10 and Galectin-9 strongly correlated with each other (Table 2), suggesting that they may arise from similar pathogenic mechanisms.

Sex and ethnicity affect cytokine levels

Given the difference in the proportion of males between ANA−HC and the ANA+ subject groups, and previous work by ourselves and others suggesting that ethnicity may affect cytokine levels, we questioned whether these demographic factors could have an impact on cytokine levels. To address this question, a multivariable analysis was performed incorporating sex and ethnicity as covariates. This analysis demonstrated that sex was a contributing factor to the variation in levels of Galectin-9 (p = 0.003) and IFN5 score (p = 0.035), whereas ethnicity affected the levels of IFN-α measured by ELISA (p < 0.0002).

The differences in cytokine levels between males and females and between Caucasian and non-Caucasian subjects for each subject group are shown in Additional file 1: Suppl. Figs. 1 and 2, respectively. Differences between male and female subjects were mostly seen in the ANA−HC group, with elevated levels of Galectin-9 and other IFN-γ-associated cytokines being significantly higher in males than females. In contrast, the differences between Caucasian and non-Caucasian subjects were predominantly restricted to the ANA+ groups, with not only IFN-α, as measured by ELISA, but also other measures of IFN-α showing variably significant increases in non-Caucasians in these groups. This finding is consistent with previous studies by ourselves and others showing elevated levels of measures of type I IFN in non-Caucasian, as compared to Caucasian, ANA+ individuals [5, 22, 23].

Serum levels of IFN-induced cytokines can be used to predict disease progression in ANA+ individuals

To examine the ability of the different cytokine measures to predict progression in ANA+ subjects lacking a SARD diagnosis, the levels of each measure were compared between clinical symptom progressors (n=13) and non-progressors (n=42), who had completed 2 years of follow-up (Fig. 2). The mean levels of Galectin-9 and CXCL-10 were significantly elevated in progressors compared to non-progressors (p = 0.0074 and p = 0.0052, respectively), and there was a trend to increased levels of IFN-α, as measured by a high-sensitivity ELISA (p = 0.059). To determine whether measurement of these cytokines could be used to predict clinical disease progression in ANA+ subjects lacking a SARD diagnosis, receiver operating characteristic curves for each cytokine measure were produced and the optimal cut-off for progression determined by calculating Youden’s index, comparing progressors and non-progressors (see Additional file 1: Suppl. Fig. 3).

Fig. 2

Cytokine levels in longitudinally followed ANA+ subjects lacking a SARD diagnosis stratified for the presence or absence of clinical progression over the subsequent 2 years. Levels of the various cytokines in clinical progressors (n = 13) and non-progressors (n = 42). Scatterplots showing the results for the IFN5 score and the cytokines, IFN-α measured by high-sensitivity ELISA, IFN-α measured by Simoa, CXCL-10, Galectin-9, and IFN-γ (all shown using logarithmic scales). Each circle represents a single subject, with the bars indicating the median for the subjects and error bars denoting the interquartile range. Significant differences between progressors and non-progressors are indicated by asterisks, ** p < 0.01

The greatest discriminative ability, as indicated by the highest area under the curve, was seen for Galectin-9, CXCL-10, and IFN-α measured by high-sensitivity ELISA, with IFN-γ and IFN-α measured by Simoa showing little or no discriminative ability. Using univariate logistic regression, we determined that ANA+ individuals with elevated levels of Galectin-9 (p = 0.008, OR = 6.34, 95% CI 1.62–24.80), CXCL-10 (p = 0.0017, OR = 9.6, 95% CI 2.34–39.42), or IFN-α (p = 0.011, OR=5.83, 95% CI 1.50–22.71) are more likely to progress within the next 2 years. To determine whether a combination of these cytokines offered an improved ability over the individual cytokine measures to predict progression, their sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were contrasted (Table 3). While the individual cytokines tended to have higher sensitivity, in general, the specificity and PPV improved with combinations of cytokines. Notably, the combination of elevated levels of CXCL-10 and IFN-α was strongly predictive of progression, with a specificity and PPV of 100%. Although the area under the curve for the IFN5 score was only slightly less than that for IFN-α measured by high-sensitivity ELISA and using the optimal cut-off for progression had a comparable specificity and PPV to IFN-α (either alone or in combination with other cytokines), it had a significantly reduced sensitivity and NPV (Additional file 1: Suppl. Table 3).

Table 3 Comparison of various combinations of cytokines as predictors of clinical progression in ANA+ subjects lacking a SARD diagnosis

We previously examined the ability of 144 autoantibodies, as measured by an autoantigen array, to predict progression and found that only elevated levels of anti-Ro52 antibodies were associated with an increased risk of development of new SARD criteria over the subsequent 2 years (PPV 46%, NPV 89%) [19]. We therefore investigated whether the measurement of this antibody improved the ability of the cytokine measures to predict progression, using our previously established cut-off. The presence of elevated levels of anti-Ro52 antibodies resulted in very modest increases in the specificity and PPV when combined with the individual cytokine measures, but offered a minimal improvement over the combinations of cytokines. Because we previously noted a moderate correlation between the IFN5 score and the presence of anti-Ro52 antibodies [19], we questioned whether elevated levels of anti-Ro52 antibodies offered the same discriminative ability between progressors and non-progressors as IFN-α. Although the presence of elevated anti-Ro52 antibody levels in tandem with elevated levels of CXCL-10 or Galectin-9 improved the sensitivity of detection of progressors, as compared to the presence of elevated levels of the same cytokine with increased IFN-α levels, it resulted in a significant decrease in the specificity and especially the PPV.

None of the progressors or non-progressors had a dense fine speckled pattern on their ANA and there was no difference in the proportion of subjects with any of the remaining ANA patterns between progressors and non-progressors.