Subjects were either recruited by local advertisement or from the outpatient clinic for treatment of obesity at the local hospital. Obesity was defined as body mass index (BMI) 30 kg/m² or greater. All subjects were healthy except for obesity and were free of medication. Participants were investigated in the morning after an overnight fast. The study was approved by the Research Ethics Committee at Karolinska Institutet, South (Stockholm) and the Institutional Research Board of INSERM and Toulouse University Hospital. Written informed consent was obtained from all participants. All the experimental procedures were conducted under the provisions of the Declaration of Helsinki.

Examined cohorts are described in Table  1. All subjects comprised women. Cohort 1 comprised five women (age 44 ± 3 yr, BMI 33 ± 11 kg/m²) that was used to evaluate in vitro secretion of AIF-1 from abdominal adipose pieces. WAT was obtained in connection with surgery for cosmetic reasons or gastric bypass surgery for treating obesity. These samples had been previously collected.

¹From each subject only one cell type could be isolated from adipose tissue. Five to six subjects were used to quantify each of the different sub-fraction of cells in adipose tissue. In total, 28 subjects were used. ²KITT was available for 46 women. ³S-Adiponectin was available for 37 women.

N. A. = Not available.

The cellular origin of AIF-1 was measured in cohort 2 comprising women undergoing abdominal dermolipectomy for cosmetic reasons (age 41 ± 2 yr, BMI 29 ± 1 kg/m²). Additional clinical data was not available for this cohort. From each subject only one cell type could be isolated from adipose tissue. Five to six subjects were used to quantify each of the different sub-fractions of cells in adipose tissue. In total, 28 subjects were used.

AIF-1 mRNA levels were quantified in subcutaneous and visceral WAT in cohort 3 comprising 12 non-obese (age 40 ± 13 yr, BMI 24 ± 2 kg/m²) and 23 obese women (age 42 ± 10 yr, BMI 44 ± 4 kg/m²). The non-obese subjects were operated for uncomplicated gallstone disease and the obese underwent anti-obesity (bariatric) surgery (BMI > 40). Subcutaneous adipose tissue from the surgical incision and omental (visceral) adipose tissue were obtained at the beginning of surgery. Only saline was given as an intravenous infusion until adipose tissue was removed.

Subcutaneous WAT AIF-1 mRNA levels were further investigated in cohort 4 comprising 10 obese women (aged 39 ± 6 yr, BMI 40 ± 6 kg/m²) investigated before and 2–4 yr after intense anti-obesity therapy with anti-obesity surgery or behavioral modification when they had reached a non-obese weight-stable state for at least 6 months (BMI 26 ± 3 kg/m²) 18 . Briefly, adjustable vertical gastric banding was used as surgical treatment. The patients were followed yearly and re-examined when body weight had reached nadir two to four years after operation and was stabilized for at least 6 months according to self report (< 1 kg increase or decrease). Behavioral modification consisted of increasing motivation, changing eating habits, enhancing physical activity and regular 3 month follow ups for a year, when re-examination took place. By then body was stabilized for at least one month according to self report (< 1 kg increase or decrease). Weight loss was 22 ± 12% (mean ± SD) in the surgery group and 14 ± 6% in the behavioural group. AIF-1 mRNA levels were related to metabolic phenotypes in cohort 5 comprising n = 96 women with a wide variation in BMI (age 39 ± 9 yr, BMI 38 ± 7 kg/m²). In cohorts 4 and 5 biopsies of the subcutaneous abdominal WAT (0.5–2 g) were obtained by needle aspiration under local anesthesia. Cohorts 1, 3, 4 and 5 comprised third-generation Caucasian Swedish women living in Stockholm. Cohort 2 comprised Caucasian French women. WAT samples were brought to the laboratory in saline, and one part was used immediately for adipocyte experiments and another part was frozen in liquid nitrogen. Unfortunately, it was not possible to recruit enough male subjects to perform these clinical studies.

Insulin tolerance test was performed as previously described 19 . Briefly, patients were examined in the hospital at 8 a.m. after an overnight fast. Following a 30–60 min rest, venous blood samples were taken for determinations of fasting levels of glucose and plasma insulin. Next, insulin was rapidly injected intravenously. After 2.5, 5, 7.5, 10, 15, 20, 25 and 30 min’s, blood glucose levels were determined and plotted in a semilogaritmic graph depicting blood glucose decrease along time. The rate constant (k) derived from this plot represented the intravenous insulin tolerance (KITT). The homeostasis model assessment for insulin resistance (HOMA_IR) was calculated as fasting serum insulin (μU/ml) x fasting plasma glucose (mmol/L)/22.5 20 .

Morphology of adipose tissue was calculated as previously described 21 . Briefly, adipocytes and stroma-vascular fraction (SVF), respectively, were isolated from adipose tissue by collagenase digestion as previously described and mean fat cell size determined 22 . Total fat mass was measured by bioimpedance. We have fitted a curve, which defines the relationship between mean adipocyte volume and fat mass. The differences between observed mean adipocyte volume in the present investigation and the expected volume as obtained from the curve fit at the corresponding level of total body fat mass were calculated separately for each of the 96 woman. The women were classified as having either hyperplastic (negative deviation) or hypertrophic (positive deviation) adipose tissue morphology relative to the estimated average value for body fat. The morphology values are quantitative: a larger absolute positive or negative value reflects more pronounced hypertrophy or hyperplasia, respectively, compared with a small value.

The stroma-vascular fraction and adipocytes were obtained after collagenase digestion and the different cell types of the stroma-vascular fraction were obtained by immunoselection and depletion as described previously 23 24 . The following cell types were identified: endothelial cells (CD34+/CD31+), fraction containing cells with capacity to differentiate into fat cells (progenitor cells; CD34+/CD31-), macrophages (CD34-/CD14+), and T lymphocytes (CD34-/CD14-/CD31+) 6 .

Adipose tissue pieces (300 mg) or 200 μL isolated cells were kept at −70°C for subsequent RNA extraction using the RNeasy minikit (QIAGEN, Hilden, Germany). RNA samples were treated with ribonuclease-free deoxyribonuclease (QIAGEN). The RNA concentration was determined using spectrophotometer and the quality of RNA was confirmed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). One microgram of RNA was reverse transcribed to cDNA using the Omniscript reverse transcriptase kit (QIAGEN) and random hexamer primers. For qPCR analysis on the distinct cell types of human adipose tissue, mRNA were extracted from macrophages, lymphocytes, adipocytes, endothelial cells, and progenitor cells isolated from human AT as previously described 6 . RNA was reverse-transcribed using the “Superscript II” kit (Invitrogen, Cergy-Pontoise, France).

Adipose tissue AIF-1 mRNA and the internal control gene 18S were quantified using the SYBR Green-based technology. Primer pairs were designed to span exon-intron boundaries. The following primers were used for specific mRNA quantification: 5’ GATGATGCTGGGCAAGAGAT-3’ and 5’ CCTTCAAATCAGGGCAACTC-3’ for AIF-1; 5’ CACATGGCCTCCAAGGAGTAAG-3’ and 5’ CCAGCAGTGAGGGTCTCTCT-3’ for 18S 25 . 5 ng of cDNA were mixed with gene-specific primers (final concentration 300 nM) and IQ SYBR green supermix (Bio-Rad Laboratories, Hercules, CA) PCR was performed with an iCycler IQ (Bio-Rad Laboratories). A direct comparative method was used for data analysis, i.e. 2^{(Ct target calibrator - Ct target sample)}/2^{(Ct 18S calibrator - Ct 18S sample)}. The PCR efficiency in all runs was close to 100% and all samples were run in duplicate.

For qPCR analysis on the distinct cell types of human adipose tissue AIF-1 primers were from Applied Biosystems (Courtaboeuf, France Hs 00610419 g1). The amplification reaction was performed in duplicate on 15 ng cDNA samples in a final volume of 20 μL in 96-well reaction plates (Applied Biosystems) in a GeneAmp 7500 detection system. All reactions were carried out under the same conditions: 50°C for 2 min, 95°C for 10 min, 40 cycles of 95°C for 15 sec and 60°C for 1 min. Results were analyzed with the GeneAmp 7500 software and all the values were normalized to the levels of 18S rRNA (Applied Biosystems).

WAT pieces (300 mg in 3 mL medium) were incubated for different periods of time as described 26 . Briefly, adipose tissue samples from cohort 1 were minced in small pieces and visible vessels and coagulation particles were removed. The remaining tissue (~300 mg) was rinsed in saline at 37°C and incubated in 3 mL Krebs-Ringer phosphate buffer (pH 7.4) supplemented with 40 g/liter of defatted bovine serum albumin and 1 g/liter of glucose. WAT samples were incubated for 3 hours at 37°C in a shaking water bath, with air as the gas phase. After 1, 2 and 3 hours, 2 mL of the medium was removed and frozen in liquid nitrogen. Medium was saved at −70°C for later determination of AIF-1 levels by ELISA (Catalog no. E92288Hu; USCNK Life Science Inc., China), according to the manufacturer’s instructions. Values were expressed as picograms per milliliter. Serum levels of adiponectin were measured using a RIA method (Linco Research, Inc., St. Charles, MO) and were expressed as micrograms per milliliter.

Unpaired or paired t test (two sided) was used to compare mRNA levels between two groups. Values are mean ± SD. We used simple regression and multiple regression with BMI as covariate to examine relationships between quantitative variables. Differences were considered as statistically significant at p < 0.05. The statistical analyses were performed using Statview software (version 5.01; SAS Institute, Cary, NC).

We first wanted to assess whether AIF-1 was secreted from human adipose tissue. Adipose tissue pieces from women in cohort 1 were incubated in vitro as described 27 and secretion of AIF-1 was measured in the conditioned medium by ELISA. AIF-1 levels accumulated in the media along the 3 h-experimental period (Figure  1) demonstrating that AIF-1 was secreted in a time dependent-manner from WAT pieces in all of the 5 experiments. This defines AIF-1 as a novel adipokine.

We next investigated the cellular source of AIF-1 in human WAT using cohort 2 (age 41 ± 2 yr, BMI 29 ± 1 kg/m²). Our results show that the stroma-vascular fraction was the major source of AIF-1 mRNA in WAT. Highest levels of AIF-1 were observed in the CD34-/CD14+ cells, which contain macrophages (P < 0.001; Figure  2). Very low levels of AIF-1 mRNA were detected in isolated adipocytes; mRNA levels in lymphocytes, endothelial and progenitor cells were in between those in adipocytes and macrophages.

WAT mRNA levels of AIF-1 were increased about two-fold in obese as compared to non-obese women from cohort 3 in both abdominal subcutaneous and visceral fat depots, (P < 0.01; Figure  3a and 3b). Neither age, nor adipose tissue depot influenced AIF-1 mRNA expression (values not shown). Furthermore, AIF-1 mRNA levels in subcutaneous WAT were normalized in obese women from cohort 4 following marked long term weight loss (P < 0.01; Figure  3c). To assess whether the association between AIF-1 mRNA levels and obesity was related to increased macrophage infiltration into adipose tissue, we wished to correlate AIF-1 levels to a macrophage marker. Integrin alpha-X (ITGAX) is a gene specific for proinflammatory macrophages 28 . We used a previously published global microarray-based gene expression profile on abdominal subcutaneous adipose tissue from obese and lean women (n = 56) 28 to show that AIF-1 and ITGAX levels were strongly correlated (r² = 0.523, P < 0.0001).

We next wanted to investigate whether AIF-1 expression in WAT was associated with clinical parameters related to insulin sensitivity in a large cohort of women (cohort 5). AIF-1 was correlated to BMI and not age in this cohort (Table  2). In multiple regression with BMI and AIF-1 mRNA levels as independent factors, we found that WAT AIF-1 expression was independently correlated with insulin resistance as determined by HOMA_IR (partial r = 0.369, P = 0.0042) in this cohort of women. In a subgroup of women insulin sensitivity as estimated by the short insulin tolerance test (KITT) and circulating levels of adiponectin had been quantified. In this group of patients WAT AIF-1 expression was inversely and independently of BMI (Table  2) correlated with insulin sensitivity as estimated by the short insulin tolerance test (KITT) (partial r = −0.396, P = 0.017, Figure  4a), and circulating levels of adiponectin (partial r = −0.334, P = 0.023, Figure  4b). Finally, WAT AIF-1 expression was independently of BMI associated with fat cell volume (partial r = 0.396, P = 0.0041; Table  2), but not adipose morphology (values not shown). In the multiple regression analysis (Table  2), BMI became non-significant as regressor in the presence of all variables except adiponectin.

¹Simple regression. ²Multiple regression with BMI together with one other variable at a time as independent variables. In the multiple regression, BMI became non-significant when examined together with KITT, fat cell volume or HOMA (partial r between 0.04 and 0.18). However it remained significant together with adiponectin (partial r = 0.44; p = 0.004) ³Insulin tolerance test was performed in patients after an overnight fast. Insulin was rapidly injected intravenously. After 2.5, 5, 7.5, 10, 15, 20, 25 and 30 min’s, blood glucose levels were determined and plotted in a semilogaritmic graph depicting blood glucose decrease along time. The rate constant (k) derived from this plot represented the intravenous insulin tolerance (KITT). ⁴Serum levels of adiponectin were measured using a RIA method. ⁵The relationship between mean adipocyte volume and total body fat determines adipocyte morphology. Delta is the difference between observed adipocyte volume in the present investigation and the expected volume (as obtained from the curve fit) at the corresponding level of total body fat mass were calculated separately for each woman. The morphology values are quantitative: a larger absolute positive or negative value (delta) reflects more pronounced hypertrophy or hyperplasia, respectively, compared with a small value, see Morphology in methods section for details. ⁶Adipocytes were isolated from adipose tissue by collagenase digestion as previously described and mean fat cell size determined as described in reference 21. ⁷The homeostasis model assessment for insulin resistance (HOMA_IR) was calculated as fasting serum insulin (μU/ml) x fasting plasma glucose (mmol/L)/22.5.