Page 20 - Read Online

Puyana et al. J Transl Genet Genom 2022;6:223-239 https://dx.doi.org/10.20517/jtgg.2021.51 Page 227

RESULTS
Constructing the obesity PRS
To build the obesity PRS, SNPs previously found to be associated with BMI and their respective proxies
were identified in the literature (n = 700) [21,26,60-76] . After genotyping quality control, 275 BMI-related SNPs
were found in our GWAS dataset using PLINK software v1.07c. Minor and major allele frequencies were
identified for each SNP using a profile option in the PLINK software.

Obesity risk alleles for each SNP were established using a linear regression model. Regression coefficients
and P values were calculated per SNP computing the risk that presenting the minor allele confers to
increasing BMI using the Assoc- function in PLINK. Only significantly related SNPs (P < 0.05) with low
linkage disequilibrium (LD < 0.8) were included in the final dataset of 35 SNPs. The PRS model was then
constructed by adding the weighted risk alleles: (1) the number of risk alleles was counted for each SNP and
multiplied by its effect size; and (2) weighted risk alleles were summed across all 35 SNPs for each patient.
RStudio was used to merge annotations files to find corresponding genes for each SNP in the obesity PRS
[Table 1].

Patient characteristics
A total of 403 breast cancer patients were analyzed: 310 were post-lumpectomy, and 93 were post-
mastectomy. Patients self-identified as HW (65%), AA (21%), and NHW (14%). The mean age at consent
2
was 55.3 ± 9.4 years (range 27-82), and the mean BMI was 28.9 ± 5.9 kg/m (range 19-63). Overall, 35% of
participants were overweight, 27% were obese class I, 8% were obese class II, and 3% were obese class III.
2
BMI differed significantly by race/ethnicity (P < 0.0001): AA had the highest mean BMI (30.98 ± 7.88 kg/m )
2
2
, followed by HW (28.79 ± 4.83 kg/m ), and NHW (26.73 ± 6.33 kg/m ) [Table 2].
Association between obesity PRS and BMI
Mean obesity PRS was evaluated by BMI category, race/ethnicity, and bariatric surgery eligibility [Table 3].
Overall mean PRS was 41.5 ± 9.93 (range 20.9-74.9; median = 39.8). The mean PRS for obese patients was
45.02 ± 10.6 and 39.3 ± 8.8 for non-obese patients (P < 0.0001). There was a significant difference in mean
obesity PRS values between each BMI category (P < 0.0001), with a corresponding dose-response increase in
obesity PRS for each increasing BMI category. When compared by race/ethnicity, the mean PRS was
significantly higher in AA (mean ± SD, 55.03 ± 7.99) compared to HW (38.27 ± 6.65) and NHW (35.94 ±
6.90) women (P < 0.0001). There was also a significantly greater mean obesity PRS for patients who were
eligible for bariatric surgery compared to those who were not (P < 0.0001).

The obesity PRS was categorized into 4 levels based on quartiles (level 1; PRS ≤ 34.3; level 2: 34.3 < PRS ≤
39.8; level 3: 39.8< PRS ≤ 47.18; and level 4: PRS > 47.18).As shown in [Table 4], patients with PRS level 4
had 3.77-fold higher odds of being obese than those with PRS level 1 (95%CI: 2.06-6.89). Conversely, the
odds of being obese among those with PRS level 2 (OR = 1.42, 95%CI: 0.76-2.65) was not significantly
different from PRS level 1. Together, our data suggest that higher mean PRS scores, particularly PRS scores
in the highest two quartiles, are significantly associated with obesity.

As illustrated in [Table 5], the mean CRP value for patients with obesity PRS level 4 (9.03 ± 17.02 mg/L) was
higher than for patients with PRS levels 1, 2, or 3, but this difference was not statistically significant.

The results of our linear regression model of the association of PRS with BMI revealed that the PRS
improved our BMI prediction by 14% [β = 0.23 (SE = 0.03), P < 0.0001] [Figure 1].